Keywords: Motion Control, Optimization and Optimal Control, Model Learning for Control
Abstract: In this paper, we introduce a novel means of control design for probabilistic movement primitives (ProMPs). ProMPs are a powerful tool to define a distribution of trajectories for robots or other dynamic systems. However, existing control methods to execute desired motions suffer from a number of drawbacks such as a reliance on linear control approaches and sensitivity to initial parameters. We propose the use of feedback linearization, quadratic programming, and multiple control barrier functions to guide a system along a trajectory within the distribution defined by a ProMP, while guaranteeing that the system state never leaves more than a desired distance from the distribution mean. This allows for better performance on nonlinear systems and offers firm stability and known bounds on the system state. Furthermore, we highlight how the proposed method may allow a designer to emphasize certain objectives that are more important than the others. A series of simulations demonstrate the efficacy of our approach and show it can run in real time.

Keywords: Motion and Path Planning
Abstract: Autonomous underwater vehicles (AUVs) are often required to reach multiple goal locations while staying within the communication range of a surface vehicle. The goals, which could be dispersed throughout the environment, are dynamically discovered by the surface vehicle as it moves along a predefined trajectory. As the goals are discovered, they are communicated to the AUV. This paper develops an efficient multi-layered planner that generates collision-free and dynamically-feasible trajectories that enable the AUV to reach as many goals as possible while always staying within the communication range of the surface vehicle. The planner relies on a roadmap to capture the connectivity of the environment in order to facilitate navigation. The high-level layer is based on discrete search to find paths over the roadmap that increase the sum of the rewards to the known goals while maintaining the range constraints. The low-level layer relies on sampling-based motion planning to expand a tree of feasible motions along paths computed by the discrete layer. These layers interact with each other to update the planned motions as new goals are dynamically discovered. Experiments using 3D environments, second-order AUV models, and an increasing number of goals, demonstrate the efficiency of the planner to solve dynamic multi-goal motion-planning problems with range constraints.

Keywords: Additive Manufacturing, Machine learning, Probability and Statistical Methods
Abstract: Geometric accuracy control is critical for precision additive manufacturing (AM). To learn and predict the shape deformation from a limited number of training products, a fabrication-aware convolution learning framework has been developed in our previous work to describe the layer-by-layer fabrication process. This work extends the convolution learning framework to broader categories of 3D geometries by constructively incorporating spherical and polyhedral shapes into a unified model. This is achieved by extending 2D cookie-cutter modeling approach to 3D case and by modeling the spatial correlation among neighboring regions. Methodologies demonstrated with real case studies show the promise of prescriptive modeling and control of complicated shape quality in AM.

Keywords: Additive Manufacturing, Hydraulic/Pneumatic Actuators, Compliant Joints and Mechanisms
Abstract: Soft robotic actuators offer a range of attractive features relative to traditional rigid robots including inherently safer human-robot interaction and robustness to unexpected or extreme loading conditions. Soft robots are challenging to design and fabricate, and most actuators are designed by trial and error and fabricated using labor-intensive multi-step casting processes. We present an integrated collection of software tools that address several limitations in the existing design and fabrication workflow for pneumatic soft actuators. We use implicit geometry functions to specify geometry and material distribution, a GUI-based software tool for interactive exploration of computational network representations of these implicit functions, and an automated tool for generating rapid simulation results of candidate designs. We prioritize seamless connectivity between all stages of the design and fabrication process, and elimination of steps that require human intervention. The software tools presented here integrate with existing capabilities for multimaterial additive manufacturing, and are also forward-compatible with emerging automated design techniques. The workflow presented here is intended as a community resource, and aimed at lowering barriers for the discovery of novel soft actuators by experts and novice users. The data gathered from human-interaction with this tool will be used by future automation tools to enable fully-automated soft actuator design based on high-level specifications.

Keywords: Cloud Computing For Automation, Simulation and Animation, Software, Middleware and Programming Environments
Abstract: Simulations play a crucial role in robotics research and education. This paper presents the OpenUAV testbed, an open-source, easy-to-use, web-based, and reproducible software system that enables students and researchers to run robotic simulations on the cloud. We have built upon our previous work and have addressed some of the educational and research challenges associated with the prior work. The critical contributions of the paper to the robotics and automation community are threefold: First, OpenUAV saves students and researchers from tedious and complicated software setups by providing web-browser-based Linux desktop sessions with standard robotics software like Gazebo, ROS, and flight autonomy stack. Second, a method for saving an individual's research work with its dependencies for the work's future reproducibility. Third, the platform provides a mechanism to support photorealistic robotics simulations by combining Unity game engine-based camera rendering and Gazebo physics. The paper addresses a research need for photorealistic simulations and describes a methodology for creating a photorealistic aquatic simulation. We also present the various academic and research use-cases of this platform to improve robotics education and research, especially during times like the COVID-19 pandemic, when virtual collaboration is necessary.

Keywords: Cyber-physical Production Systems and Industry 4.0, Machine learning, Foundations of Automation
Abstract: Optimization-based falsification, or search-based testing, is a method of automatic test generation for Cyber-Physical System (CPS) safety evaluation. CPS safety evaluation is guided by high level system requirements that are expressed in Signal Temporal Logic (STL). Trajectories from executed CPS simulations are evaluated against STL requirements using satisfaction robustness as a quantitative metric. In particular, robustness is the distance metric between the simulated system trajectory, associated to a specific input, and the known unsafe set, i.e., regions of the search space that violate the requirements. The problem of identifying a violation can be formulated as an optimization, where inputs that minimize the robustness function are of interest. In fact, an input falsifies a requirement if the associated robustness is negative.

In this work, specifically, we consider the case where multiple requirements determine the unsafe set. Due to the computational burden of executing CPS simulations, practitioners often test all system requirements simultaneously by combining the requirement components and obtaining so-called "conjunctive requirements". Conjunctive requirements can challenge optimization-based falsification approaches due to the fact that the robustness function may "mask" the contributions of individual conjunctive requirement components. We propose a new algorithm, minimum Bayesian optimization (minBO), that deals with this problem by considering the contributions of each component of the conjunctive requirement. We show the advantages of the minBO optimization algorithm when applied to general non-linear non-convex optimization problems as well as when applied to realistic falsification applications.

Keywords: Renewable Energy Sources, Power and Energy Systems automation, Sustainable Production and Service Automation
Abstract: This work proposes a UPS (uninterrupted power supply) system for voltage sag immunity in IC manufacturing based on lithium iron phosphate battery energy storage technology and designs a lithium iron phosphate battery energy storage system and its battery management system. Oriented to the requirements of semiconductor manufacturing which includes high-voltage, high-current, high-power and high-rate performance, this work uses lithium iron phosphate battery modules as energy storage modules and establishes a hierarchical centralized battery management system to achieve “one-to-many” management mode of each single battery. Combined with the flexible sampling process of the battery management system, the system can accurately capture the abnormalities of the power grid such as devices’ voltage fluctuations and sags and provides an optimized power configuration plan. Based on high-precision battery managing algorithms, the system realizes functions of temperature protection, over-current protection, voltage protection and battery balancing. The performance tests of the discharge module, communication, current acquisition, the BCU and the UPS power system are performed for the UPS system. The test results show that the various indicators of the system meet the design requirements, and the system can solve the problems of voltage sag and power supply interruption, and can provide uninterrupted, stable and reliable AC power for precision IC manufacturing.

Keywords: Semiconductor Manufacturing, Petri Nets for Automation Control, Reinforcement
Abstract: (This paper has been recently accepted for TASE. DOI: 10.1109/TASE.2020.3047924)

A dual-gripper robotic cell consists of multiple processing machines and one material handling robot, which can perform an unloading or a loading task one at a time but can hold two parts at the same time. We address a scheduling problem of the robotic cell that determines a robot task sequence when two part types are processed in a different set of machines and all machines have variable processing times within a given interval. The objective is to minimize the makespan. This study proposes a learning-based method, i.e., a reinforcement learning (RL) approach, for the first time, to address a dual-gripper robotic cell scheduling problem. The problem is modeled with a Petri net, a graphical and mathematical modeling tool, which is used as an environment in RL. The states, actions, and rewards are defined by using flow shop scheduling properties, features from a Petri net, and knowledge from previous studies of scheduling robotized tools. Then, the RL approach is compared to the firstin-first-out (FIFO) rule, which is generally used in practice, a swap sequence, which is widely used for cyclic scheduling of dual-gripper robotic cells, and a lower bound. The extensive experiments show that the proposed method performs better than FIFO and the swap sequence; moreover, the gap between the makespan of the proposed method and the lower bound is not large.

Keywords: Failure Detection and Recovery, Semiconductor Manufacturing, AI-Based Methods
Abstract: This work deals with a change point detection problem of semiconductor manufacturing equipment, especially for a deposition process with real data. We propose a l1 trend filtering-based change point detection method. It shows an adequate true detection rate while effectively reducing false alarms compared to other methods. We further develop a pressure prediction model to improve the performance of the change point detection algorithm.

Keywords: Semiconductor Manufacturing, Intelligent and Flexible Manufacturing
Abstract: In this paper, we are interested in supporting short-term decision-making in terms of cycle time (fabrication time) in semiconductor manufacturing. In particular, we want to prioritize re-qualification decisions. Instead of resorting to simulation models, which can be prohibitive to develop and maintain, we evaluate two simple closed-form solutions to forecast cycle times on the short term.

Keywords: Semiconductor Manufacturing, Factory Automation, Manufacturing, Maintenance and Supply Chains
Abstract: The overhead hoist transport (OHT) system is the primary automated material handling system in modern semiconductor fabrication facilities. In an OHT system, a vehicle unit travels on tracks, delivering lots between machines. This motion induces mechanical stresses and vibrations in the track components. In this industry paper, we propose an abnormality detection algorithm and method for inspecting OHT tracks. Detecting hazardous objects on the tracks, such as chips, small bolts, and ceiling debris, is important to guarantee the safe and reliable operation of OHT systems. We present an autoeconder algorithm to identify track abnormality. In addition, we report the current progress of an industry--academia collaborative project to solve the aforementioned problems and to open the discussion for further development.

Keywords: Cyber-physical Production Systems and Industry 4.0
Abstract: The effective control of the welding quality can ensure the overall performance of the product. For the geometric feature online inspection system of welding process, this paper used Digital Twin technology to realize the mapping integration between physical and virtual workshop. The paper also proposed a control chart pattern recognition method based on distance mode profile of time series and convolutional neural network, along with an ensemble learning model for the decision of fixture adjustment. The experiment results show that the integration of digital twin and machine learning provides a feasible way for real-time monitoring and accurate control of welding quality.

Keywords: Cyber-physical Production Systems and Industry 4.0, Manufacturing, Maintenance and Supply Chains, Petri Nets for Automation Control
Abstract: Industry 4.0 determined the emergence of technologies which allow for data-based production planning and control approaches. Digital twins can be used to take decisions based on the current system state. Hence, their performance strictly depends on the capability to correctly represent their physical counterparts at any time. The development of digital twins for manufacturing systems can be significantly accelerated by automated model generation techniques. However, production systems including assembly stations suffer from event records with multiple part identifiers, resulting in the wrong finding of the system structure. In this paper, we define the problem of the proper discovery of assembly operations. Then, we describe an algorithm to generate a complete digital model exploiting the new concept of object-centric process mining. In a case study, a flow shop including assembly stations is successfully discovered, allowing for the automated building of a simulation model with the proper logical behavior.

Keywords: Assembly, Process Control, Hybrid Strategy of Intelligent Manufacturing
Abstract: With the increasing complexity of mechanical products, assembly, as the last step of product manufacturing, is becoming more and more complicated. Therefore, the problems in the actual assembly process are difficult to be considered at the beginning of the design. In addition, the measured information of the assembly site is also very important for the adjustment of the assembly process design. The emergence of digital twin solves this problem well, and provides the possibility for the transmission and feedback of assembly site information. However, the current digital twin data storage side mainly uses the traditional database, which leads to data redundancy. In recent years, the popular knowledge graph has powerful knowledge representation and reasoning ability, which can solve the above problems. In this paper, a digital twin system structure of assembly process based on knowledge graph is proposed, which is used to record actual process data. It provides a possibility for the combination of knowledge graph technology and digital twin technology.

Keywords: Machine learning, Factory Automation
Abstract: This paper addresses the achievements of the QU4LITY project, funded by the Swiss Innovation Agency. Within this research project, the automation back-bone developed for the SUPSI Mini Factory [1], coupled with adaptive Machine Learning (ML) algorithms, is exploited to test numerous families of bearings. The SUPSI Mini Factory is associated with an evolving digital profile, a digital twin, based on constant synchronization through IoT devices, that encompasses a massive, real-time, real- world data, gathered from the different sensors interfaced via OPC-UA protocol. This representation feeds a ML algorithm capable of 1) detecting defective bearings and 2) continually tuning the quality testing process parameters based on the analysis performed on the gathered data. Specifically, the identification of defective bearings is performed by a voting classifier fed by statistical metrics measured from the collected experiments. Our approach also aims at continually learning from the tests performed with the cell: the ML algorithm is tuned and adjusted every time a new set of tests is executed. Tests show that the proposed ML classifier accurately distinguishes damaged and uncorrupted bearings (accuracy ∼100%, precision ∼100%, recall ∼100%, TNR∼100%). The digitalization of the quality control process is furtherly enriched by the development of a data aggregator platform. The platform fetches, cleans and aggregates data from every different machine belonging to the quality cell, enabling, also remotely, the pattern-tracking, the significant KPIs displaying and the comprehensive system behavior monitoring. The platform features also as HMI for the operator, who can monitor parameters and schedule new test campaigns.

Keywords: Intelligent and Flexible Manufacturing, Learning and Adaptive Systems, Hybrid Strategy of Intelligent Manufacturing
Abstract: Digital twin technology is the core promoter of intelligent autonomous machining systems, which has been gradually explored and applied in the machining field. The digital twin system can observe, analyze, and control the machining process in real-time by creating high fidelity virtual entity of the physical entity. However, the current digital twin system is usually customized for specific scenarios, which lacks sufficient robustness. Remodeling may lead to poor modeling effects and low modeling efficiency due to insufficient data. This paper explores an adaptive evolution mechanism of the digital twin decision-making model through incremental learning and transfer learning. Besides, the specific implementation method is concluded.

Keywords: Discrete Event Dynamic Automation Systems, Formal Methods in Robotics and Automation, Software, Middleware and Programming Environments
Abstract: Given a model of an uncontrolled system and a requirement specification, a supervisory controller can be synthesized so that the system under control adheres to the requirements. There are several ways in which informal behavioral safety requirements can be formalized, one of which is using mutual state exclusion requirements. In current implementations of the supervisor synthesis algorithm, synthesis may be inefficient when mutual state exclusion requirements are used. We propose a method to efficiently enforce these requirements in supervisor synthesis. We consider symbolic supervisor synthesis, where Binary Decision Diagrams are used to represent the system. The efficiency of the proposed method is evaluated by means of an industrial and academic case study.

Keywords: Discrete Event Dynamic Automation Systems, Intelligent Transportation Systems, Domain-specific Software and Software Engineering
Abstract: Traffic management systems are used to efficiently use the available highway infrastructure and to improve traffic safety. For example, traffic on the Dutch highway is measured and controlled using around 6000 roadside units (RSUs). Despite the many similarities between these RSUs, variations exist in the number of components and configuration of an RSU. This results in a labor-intensive, error-prone design process of supervisory controllers for all possible configurations. In this paper, supervisory control theory is applied to RSUs. It allows designers to create a correct-by-construction supervisor, using models of the system and the control requirements. Due to the variations between RSUs, it is not feasible to make a model for each RSU individually. Therefore, a configurator is proposed which allows for generation of models for any given RSU configuration. The generated models are then used to create a supervisor that adheres to the specified requirements. The use of a configurator reduces the possibility of man-made errors and allows to easily create models for a large number of systems. As a proof of concept, a model for a representative RSU is generated.

Keywords: Model Learning for Control, Discrete Event Dynamic Automation Systems, Formal Methods in Robotics and Automation
Abstract: A tool, MIDES, for automatic learning of models and supervisors for discrete event systems is presented. The tool interfaces with a simulation of the target system to learn a behavioral model through interaction. There are several different algorithms to choose from depending on the intended outcome. Moreover, given a set of specifications, the tool learns a supervisor that can help ensure the controlled system guarantees the specifications. Furthermore, the state-space explosion problem is addressed by learning a modular supervisor. In this paper, we introduce the tool, its interfaces, and algorithms. We demonstrate the usefulness through several case studies.

Keywords: Discrete Event Dynamic Automation Systems, Logistics, Planning, Scheduling and Coordination
Abstract: In this paper we consider a problem of controlling multi-agent discrete-event systems to serve tasks dynamically appearing in the environment, where the tasks have different priorities. To address this problem, we propose an effective online supervisory control approach which uses different queues to store tasks of different priorities, and assigns agents to serve tasks in the order of their priorities. Moreover, to prevent lower priority tasks from being unserved due to constantly incoming higher-priority ones, a timer is further associated with each task; once the timer of a task ticks down to zero, regardless of the task’s priority, it will be moved to a special queue with the highest priority to be served next. We then apply this online control scheme to model and control a warehouse automation system using multiple mobile robots with prioritized tasks; the effectiveness of this scheme is demonstrated on a case study.

Keywords: Petri Nets for Automation Control, Discrete Event Dynamic Automation Systems
Abstract: This paper concerns the security analysis of discrete event systems modeled with a particular class of synchronized Petri nets that include output functions, called Output Synchonized Petri nets. Such a formalism is suitable and tractable to represent a large variety of cyber-physical systems. In particular, we study here cyber-attacks that aim to drive the system from a given normal state to forbidden state. We assume that the attacker has a certain credit to insert and delete input and output events, depending on its own objectives. The proposed analysis aims to evaluate the costs of stealthy attacks on the controlled system depending on the objective of the controller, the structure of the system and the cost of the malicious actions.

Keywords: Cellular and Modular Robots
Abstract: Snakes are a remarkable evolutionary success story. Numerous snake-inspired robots have been proposed over the years. Soft robotic snakes (SRS), with their continuous and smooth bending capability, can better mimic their biological counterparts’ unique characteristics. Prior SRSs are limited to planar operation with a limited number of planar gaits. We propose a novel SRS with spatial bending ability and investigate snake locomotion gaits beyond the planar gaits of the state-of-the-art systems. We derive a complete floating-base kinematic model of the SRS and use the model to derive joint-space trajectories for serpentine and inward/outward rolling locomotion gaits. These gaits are experimentally validated under varying frequency and amplitude of gait cycles. The results qualitatively and quantitatively validate the proposed SRSs’ ability to leverage spatial bending to achieve locomotion gaits not possible with current SRS designs.

Keywords: Big Data in Robotics and Automation, Process Control, Probability and Statistical Methods
Abstract: Additive Manufacturing (AM) enables the direct production of complex geometries from computer-aided designs (CAD)in a layer-by-layer manner, whereby minute printing errors in one layer can manifest significant defects in the final part. In-situ quality monitoring and control are currently limited for AM processes and cause low repeatability. Recently. advanced imaging is increasingly invested in AM and leads to the proliferation of layerwise imaging data, which provides an opportunity to transform quality control of AM from post-build inspection to in-situ quality monitoring. However, existing methodologies for in-situ inspection primarily focus on key characteristics of image profiles that tend to be limited in the ability to analyze the variance components, as well as root causes and failure patterns that are critical to process improvement. This paper presents an Additive Gaussian Process with dependent layerwise correlation (AGP-D) to model the spatio-temporal correlation of layerwise imaging data for AM quality monitoring. The AGP-D consists of three independent GP modules. The first GP approximates the base profile, whereas the second and third GP capture the correlation within the same layer and among layers, respectively. Based on posterior predictions of new layers, Hotelling T^2 and generalized likelihood ratio (GLR) control tests are formulated to detect process shifts in the newly fabricated layer and analyze root causes. The proposed methodology is evaluated and validated using both simulation data and real-world case study of a cylinder build fabricated by a laser powder bed fusion (LPBF) machine. Experimental results show the proposed AGP-D is effective for real-time modeling and monitoring of layerwise-correlated imaging data.

Keywords: Robotics and Automation in Life Sciences
Abstract: Water visibility is a critical matter for underwater robots during operations, both for getting clear views of the environment, and as a calm and disturbance-free operating region for the manipulators to perform sampling or other operations. In reality, calm and clear water is not only restricted by the natural conditions, but also often hindered by the propeller operations from the robots themselves. Frequent gesture adjustments during manipulator operations are particularly stir-inducing from state-of-the-art underwater manipulative robots. In this work, tackling the flow disturbance issue,

a novel underwater robotic platform was proposed with a jellyfish-inspired holonomic platform driven by propellers, a flow-deflection middle layer as an adjustable isolation, and a soft-robotic manipulator mounted below the bottom for operations. Compared with state-of-the-art works, the proposed platform achieved holonomic underwater locomotion with the vertical main direction having

significantly reduced flow resistance; the flow-deflection layer could create a flow-calm region of 18.5 times larger underneath for the manipulator operations. A prototype robot was fabricated, and tested in

closed- and open-water conditions. Results were compared to flow-simulation results with good agreements, verifying that the proposed jellyfish-inspired robotic concept was effective in both asymmetric underwater locomotion and reducing water disturbances for underwater manipulator operations.

Keywords: Software Architecture for Robotic and Automation
Abstract: This paper presents a self-contained system for the robust utilization of aerial robots in the autonomous exploration of cave environments to help human explorers, first responders, and speleologists. The proposed system is generally applicable to an arbitrary exploration task within an unknown and unstructured subterranean environment and interconnects crucial robotic subsystems providing full autonomy of the robots. Such subsystems primarily include mapping, path and trajectory planning, localization, control, and decision making. Due to the diversity, complexity, and structural uncertainty of natural cave environments, the proposed system allows for the possible use of any arbitrary exploration strategy for a single robot, as well as for a cooperating team. A multi-robot cooperation strategy maximizing the limited flight time of each aerial robot is proposed for exploration and search & rescue scenarios where the homing of all deployed robots back to an initial location is not required. The entire system is validated in a comprehensive experimental analysis comprising of hours of flight time in a real-world cave environment, as well as by hundreds of hours within a state-of-the-art virtual testbed that was developed for the DARPA Subterranean Challenge robotic competition. Among others, experimental results include multiple real-world exploration flights traveling over 470 m on a single battery in a demanding unknown cave environment.

Keywords: Embedded Systems for Robotic and Automation, Industrial Robots, Software Architecture for Robotic and Automation
Abstract: Soft robots are a group of robots made from highly compliant materials. Compared to their rigid counterparts, soft robots show better mobility and stronger adaptability to the environment. Specifically, soft quadruped robots, using four soft actuators as robot legs, have become a popular design, and are proven to be feasible and reliable in performing soft locomotion. Traditional soft quadruped robots are actuated using pneumatic forces which require high pressure sources. Additionally, the travel range of these robots are limited. This paper proposes a new design of soft quadruped robot that is fully actuated with electric motors. The main components of the robot are 3D printed and can be easily assembled. The four soft legs of the robot are modeled with experimental data and a closed loop controller is developed to regulate the deformation and motion. The legs can perform complex movement which enables the quadruped robot to move with different gaits.

Keywords: Control Architectures and Programming, Discrete Event Dynamic Automation Systems, Software, Middleware and Programming Environments
Abstract: Highly-dynamic robotic systems, such as hopping robots, require light, computationally and energy efficient on-board units for control. This paper presents such a computational unit together with a software architecture for achieving high-performance behaviors, such as balancing and hopping. These demanding behaviors require accurate dynamic calculations, high-bandwidth control, and fast real-time state estimation. The proposed system consists of cheap and off-the-shelf electronics that are detailed in this paper. The effectiveness of the presented approach is validated on a balancing machine called Tippy, which is able to achieve fast tracking of command signals while balancing. The experimental results of this paper demonstrate that reliable real-time software for demanding high-performance robotic applications, which require fast control loops and intensive calculations, can be achieved with light, low cost and energy efficient components, which can empower the widespread use and experimentation of high-performance robots worldwide.

Keywords: Optimization and Optimal Control, Inventory Management, Reinforcement
Abstract: Because different items of the perishable products have different deterioration extents in an inventory system, the deterioration process of these perishable products cannot be captured accurately by a single number like deterioration rate. Instead, it is more appropriate to capture their deterioration process on different freshness levels by employing the freshness transition function. Especially, time-temperature indicator (TTI) technology can detect the freshness deterioration process accurately. Therefore, the freshness transition function of perishable products can be constructed based on the data collected by the TTIs. The accuracy of a freshness transition function depends on the detection quality of the TTIs deployed in the perishable inventory system and affects the performance of the pricing decision, because this decision is made based on the observation accuracy of the products’ freshness. In this research, we develop the method of optimal pricing decision to obtain the maximum total profit based on the freshness transition function. This method is implemented in three steps: constructing the freshness transition function, designing the pricing optimization policy based on Deep Q-network method, and analyzing the impact of the accuracy of the freshness transition function on the performance of the pricing decision. Finally, we conduct some numerical experiments to examine the performance of the proposed optimal pricing method based on the freshness transition function and found that (1) the retailers should increase the price to obtain more total profit within a sale cycle; (2) the accuracy of the freshness transition function has great influence on the proposed pricing optimization policy.

Keywords: Optimization and Optimal Control, Task Planning
Abstract: Highly maneuverable, over-actuated aerial robots have gained increasing interest in various inspection applications. However, since these systems carry expensive equipment and must operate in the vicinity of humans, their fail-safe operation is paramount. Hence, we propose a centralized nonlinear model predictive control (NMPC) method to facilitate fault-tolerant control (FTC) of an over-actuated quadrotor against a propeller failure. Thanks to the novel mechanical design, the hyperdynamic quadrotor can independently command and control all 6-degrees-of-freedom (DoFs). Additionally, the underlying reconfigurability of the designed NMPC makes it appropriate for both nominal and faulty operations. Moreover, the centralized nature of the control framework fully exploits the actuator redundancy, thereby ensuring complete system control without losing any DoF. The efficacy of the proposed FTC framework is elaborated via intensive simulations utilizing a realistic model over two different trajectories. From the considered sequential failure cases, it is shown that – even with fault detection delay up to 1s – the aerial robot satisfactorily tracks the reference trajectory.

Keywords: Optimization and Optimal Control, Planning, Scheduling and Coordination, Power and Energy Systems automation
Abstract: In order to overcome the shortcoming that the dual distributed sub-gradient optimization methods need to construct a feasible solution, a novel distributed sub-gradient optimization method based on primal decomposition is proposed in this paper and used to solve the joint dynamic economic dispatch (JDED) problem of multi-area interconnected power systems (MAIPSs). Firstly, the centralized optimization model is established and decomposed into multiple independent local areas’ optimization and a global coordinator’s optimization by splitting area power grids and cross-area tie-lines. Moreover, the slack variables and corresponding penalties are introduced into the local optimization to ensure feasibility and optimality. Secondly, a distributed sub-gradient optimization method is proposed to solve the decomposed model, in which the sub-gradient is calculated by using the dual multipliers from local optimization. Furthermore, in order to get better convergence, the heuristic updating rules for step size and penalty factor are designed. Finally, the numerical tests are carried out on two interconnected systems of different scales, and results show that the proposed method can obtain a good feasible solution directly and has high computational efficiency.

Keywords: Optimization and Optimal Control, Big-Data and Data Mining
Abstract: Room allocation is a challenging task in detention centers since lots of related detainees need to be held separately with limited rooms. It is extremely difficult and risky to allocate rooms manually, especially for organized crime groups with close connections. To tackle this problem, we develop an intelligent room allocation system for detention centers to provide optimized room allocation schemes automatically. We first formalize the detention room allocation problem as inverse graph partitioning, which can measure the quality of room allocation schemes. Then, we propose two heuristic algorithms to achieve the global optimization and local optimization of detention room allocation. Experiment results on real-world datasets show that the proposed algorithms significantly outperform manual allocation and suggest that the system is of great practical application value.

Keywords: Calibration and Identification, Optimization and Optimal Control
Abstract: The impacts caused by floods years from years affect the planet's surface, thus grows the relevance for studies of flood prevention. This field aims to minimize, or even avoid, consequences resulting by these natural events. This work seeks to contribute to this area by proposing a model predictive control of dam floodgates in a hydrographic basin. The control strategy is based on hydrological models of forecasting models, adapted to the proposal, and obtained with systems identification techniques. The developed methodologyis applied to a real-world case, the Itajaí River Basin in Brazil, to assess its effectiveness and illustrate the results obtained.

Keywords: Optimization and Optimal Control, Industrial and Service Robotics, Manipulation Planning
Abstract: Grasp planning for multi-fingered hands is still challenging due to the high nonlinear quality metrics, the high dimensionality of hand posture configuration, and complex object shapes. Analytical-based grasp planning algorithms formulate the grasping problem as a constraint optimization problem using advanced convex optimization solvers. However, these are not guaranteed to find a globally optimal solution. Data-driven based algorithms utilize machine learning algorithm frameworks to learn the grasp policy using enormous training data sets. This paper presents a new approach for grasp generation by formulating a global optimization problem with Bayesian optimization. Furthermore, we parameterize the object shape utilizing the Gaussian Process Implicit Surface~(GPIS) to integrate the object shape information into the optimization process. Moreover, a chart defined on the object surface is used to refine the palm pose locally. We introduced a dual optimization stage to optimize the palm pose and contact points separately. We further extend the Bayesian optimization by utilizing the alternating direction method of multipliers(ADMM) to eliminate contact optimization constraints. We conduct the experiments in the graspit! Simulator that demonstrates the effectiveness of this approach quantitatively and qualitatively. Our approach achieves a 95% success rate on various common objects with diverse shapes, scales, and weights.

Keywords: Big data Analytics for Large-scale Energy Systems, Modelling, Simulation and Validation of Cyber-physical Energy Systems, Deep Learning in Robotics and Automation
Abstract: The cascading failure is a typical failure propagation process which can cause significant consequence to the power system. It can be triggered by the vulnerable set composed of combinations of transmission lines with specific failures. So it is of great significance to identify the vulnerable set. In this paper, we propose an identification model for the vulnerable set under deep learning framework. The main part of the model consists of Autoencoder and classification network for reducing dimensionality and identifying vulnerable set respectively. The model is trained by the data generated from cascading failure simulation platform. We conduct experiments on IEEE 30-Bus and 200-Bus systems with different initial failures to validate the identification and generalization capability. And the time consumption is also discussed to demonstrate the efficiency of the model. All of indicators prove that the model is capable of identifying the vulnerable set effectively.

Keywords: Sustainability and Green Automation, Process Control, Deep Learning in Robotics and Automation
Abstract: Due to inherent irregularities in recyclable materials, sorting valuable metals (e.g., aluminum and copper) via mechanical means is a difficult task resisting full automation. A particularly hard challenge in the domain is the separation of scrap metal pieces with physically attached impurities, which is further complicated by variations in different batches of recyclable materials. In this work, leveraging the latest development in machine learning and robot learning, we develop an image-based sorting system for tackling this challenging task. In addition to delivering a highly accurate deep learning model for reliably distinguishing pure scrap pieces from pieces containing impurities with over 95% precision/recall, we further automate the process of sample preparation, data acquisition/labeling/analysis, and machine learning model training.

Keywords: Deep Learning in Robotics and Automation, Planning, Scheduling and Coordination, Reinforcement
Abstract: Motivated towards performing missions in unstructured environments using a group of robots, this paper presents a reinforcement learning-based strategy for multiple quadrupedal robots executing collaborative manipulation tasks. By taking target position, velocity tracking, and height adjustment into account, we demonstrate that the proposed strategy enables four quadrupedal robots manipulating a payload to walk at desired linear and angular velocities, as well as over challenging terrain. The learned policy is robust to variations of payload mass and can be parameterized by different commanded velocities.

Keywords: Deep Learning Methods, Planning, Scheduling and Coordination, Industrial Robots
Abstract: In this paper, a method of motion generation for articulated manipulators is proposed. In this method, motion sequence is generated based on past motion experience. The rough procedure is as follows. Various reaching motions are performed by the manipulator in advance, and then, these experiences are stored by the manipulator as end-effector poses, joint angles, environment map, and so on. Then, the manipulator generates a new motion sequence by efficiently using the past experiences when a new target end-effector pose is given. The proposed method comprises two phases: similar experience search and motion modification. In the latter phase, a neural network is utilized to calculate the amount of modification movement. It ensures a certain level of trajectory reproducibility. The effectiveness of the proposed method is proven via simulation considering planar and 3D manipulators.

Keywords: Deep Learning in Robotics and Automation, Computer Vision in Automation, Factory Automation
Abstract: Owing to human labor shortages, the automation of labor-intensive manual waste-sorting is needed. Deep-learning-based object detector can effectively detect different types of waste items. Our proposed robotic dataset-collection system enables automatically collecting object images to quickly train a deep neural-network model. The major issue in this study is the removal of differences in the appearance of target objects from two scenes: one scene for the dataset collection and the other scene for the waste sorting in a recycling factory. If the differences exist, the performance of a trained waste detector would be decreased. We address differences in illumination and background by applying object scaling, histogram matching with histogram equalization, and background synthesis to the target-object images. With the proposed method, we could achieve higher performance than the methods that do not consider the differences.

Keywords: Deep Learning in Robotics and Automation, Data fusion, Diagnosis and Prognostics
Abstract: To monitor the degradation status of units and prevent unexpected failures in engineering systems, Health Index (HI)-based data fusion technologies have been rapidly developed by combining multiple sensor signals, which are helpful to understand the degradation processes of units and predict their Remaining Useful Lifetime (RUL). While promising, existing HI-based data fusion models for degradation modeling are still limited due to the restrictive assumptions made during the fusion or the degradation modeling processes, e.g., assuming the fusion model as a linear or kernel-based function from multiple sensor signals, or modeling the degradation process by a pre-selected basis function. Such assumptions are often invalid in industrial practice and may fail to accurately characterize the complicated relations between multiple sensor signals and the underlying degradation process. To address the issue, this paper proposes a generic indirect deep learning method that constructs an HI by combining multiple sensor signals to better characterize the degradation process. In particular, our innovative idea is to seamlessly integrate a Deep Neural Network (DNN) and a Long Short Term Memory (LSTM) model to construct the HI by fusing multiple sensor signals and characterize the degradation process, which can be applied to the degradation modeling of various engineering systems. Domain knowledge including the concept of failure threshold and monotonicity of the degradation process is also considered to enhance the interpretability of the proposed method. Simulation studies and a case study on the degradation of aircraft gas turbine engines are presented to validate the performance of the proposed method.

Keywords: Motion and Path Planning, Optimization and Optimal Control, Industrial and Service Robotics
Abstract: In recent years, a number of trajectory optimization algorithms have been proposed and established for motion planning of robot manipulators in complex, but static, predefined environments. To enable reactive motion planning under uncertain conditions caused, for example, by moving obstacles, this paper proposes a formulation of the trajectory optimization problem that is tailored for model predictive control. The proposed algorithmic solution leverages off-the-shelf computational tools for nonlinear model predictive control, optimization, and collision checking. In addition, a motion planning paradigm is introduced to allow for online collision-free motion when following a joystick command. The approach is validated in the context of an industrial pick-and-place application using MATLAB and a Kinova robot manipulator, both in simulation and with actual hardware.

Keywords: Motion and Path Planning, Optimization and Optimal Control, Formal Methods in Robotics and Automation
Abstract: An optimized static gait that combines pose optimization, motion sequence optimization, and a novel high-level planning algorithm is proposed for quadruped robots to walk on stairs. Firstly, an optimized pose is determined for the robot to stand on stairs statically. Then, a climbing gait cycle with an optimized motion sequence is presented, which takes the robot from one position and pose to another position and pose. Finally, a high-level planning algorithm is proposed to adjust the step length in each gait cycle to enable the robot to safely walk along the stairs. The proposed static gait maximizes the stair-climbing capability significantly while still guaranteeing walking safety, which provides a general solution for quadruped robots to walk on stairs of different sizes. Several simulations in V-REP are presented to evaluate the effectiveness of the optimized static gait generation technique in improving the stair-climbing capability. Compared to other quadruped robots developed recently, the robot tested in this paper can walk on a regular staircase with a rise of 20 cm and an inclination of 37.6°, and can also climb over a few steep narrow steps with a rise of 18 cm and a run of 5 cm.

Keywords: Motion and Path Planning, Motion Control
Abstract: Underactuated mechanical systems are well-known for the challenges they pose to motion planning and control. In particular, the motion planner for these systems must take into account their dynamics, while the controller has to stabilize all degrees of freedom with a fewer number of actuators. One common way to overcome these hurdles is by employing the concept of virtual (holonomic) constraints for planning and transversal stabilization for control. This paper investigates the use of piecewise virtual constraints to improve motion planning for the systems with one degree of underactuation and demonstrates that periodic motions planned with piecewise constraints can be stabilized using existing tools for orbital stabilization. We have conducted extensive experimental evaluation of the algorithm on a practical pendubot robot, and the results suggest that the proposed method yields considerably smaller cost function values than other approaches, while exhibiting faster convergence.

Keywords: Motion and Path Planning
Abstract: This paper proposes the use of antenna-based aerial inspections. We develop an approach that enables an unmanned aerial vehicle (UAV) equipped with a microwave remote sensing (MRS) system to effectively scan nonflat terrains. The approach relies on probabilistic sampling and phase shifting to generate a set of waypoints that collectively cover the area of interest. A short tour is also generated that enables the UAV to visit the waypoints in an appropriate order that reduces the overall distance traveled. Experiments with models of real nonflat terrains demonstrate the effectiveness of the approach.

Keywords: Motion and Path Planning, Collision Avoidance, Optimization and Optimal Control
Abstract: This paper presents an approach to collision-free, long-range trajectory generation for a mobile robot in an industrial environment with static and dynamic obstacles. For the long range planning a visibility graph together with A* is used to find a collision-free path with respect to the static obstacles. This path is used as a reference path to the trajectory planning algorithm that in addition handles dynamic obstacles while complying with the robot dynamics and constraints. A Nonlinear Model Predictive Control (NMPC) solver generates a collision-free trajectory by staying close to the initial path but at the same time obeying all constraints. The NMPC problem is solved efficiently by leveraging the new numerical optimization method Proximal Averaged Newton for Optimal Control (PANOC). The algorithm was evaluated by simulation in various environments and successfully generated feasible trajectories spanning hundreds of meters in a tractable time frame.

Keywords: Motion and Path Planning, Optimization and Optimal Control, Formal Methods in Robotics and Automation
Abstract: In the research of quadruped robots, stability is a very important consideration for gait design. When the robots have symmetrical structure, stability can be easily guaranteed. However, when the robots are carrying some additional devices or payloads unevenly, the position of the center of gravity (COG) may deviate from the geometrical center, which makes it a challenging task to guarantee stability. To handle this, it is of great significance to improve the stability margin during gait design. To this end, a smooth static walking gait with the maximum stability margin is developed in this paper. An algorithm of COG trajectory optimization based on the lemniscate of Gerono is proposed. The advantage of this algorithm is that the COG trajectory is smooth and continuous at any order, which avoids abrupt changes in velocity or acceleration of the robot during walking. The two parameters in the lemniscate are the main tuning parameters. According to the size of the robot, the algorithm can automatically calculate the optimal parameters (adjust the shape of the Gerono lemniscate curve) and balance the relationship between the step size and the stability margin during the robot movement. Simulation results demonstrate the effectiveness of the proposed method, and we use a mass block experiment to prove the insensitivity of the gait algorithm to the position of the COG.

Keywords: Modelling, Simulation and Optimization in Healthcare, Health Care Management
Abstract: Health care systems are at the front line to fight the COVID-19 pandemic. Emergent questions for each hospital are how many general ward and intensive care unit beds are needed, and additionally, how to optimally allocate these resources during demand surge to effectively save lives. However, hospital pandemic preparedness has been hampered by a lack of sufficiently specific planning guidelines. In this paper, we developed a hybrid computer simulation approach, with a system dynamic model to predict COVID-19 cases and a discrete-event simulation to evaluate hospital bed utilization and subsequently determine bed allocations. Two control policies, the type-dependent admission control policy and the early step-down policy, based on patient risk profiling, were proposed to lower the overall death rate of the patient population in need of intensive care. The model was validated using historical patient census data from the University of Florida Health Jacksonville, Jacksonville, FL. The allocation of hospital beds to low-risk and high-risk arrival patients to achieve the goal of reducing the death rate, while helping a maximum number of patients to recover was discussed. This decision support tool is tailored to a given hospital setting of interest and is generalizable to other hospitals to tackle the pandemic planning challenge.

Keywords: Modelling, Simulation and Optimization in Healthcare, Clinical and Operational Decision Support, Health Care Management
Abstract: Proton therapy is a highly targeted radiation treatment for tumors. In a typical proton therapy system, multiple gantries share a beam accelerator. A patient is often called back to a gantry immediately after the gantry becomes available, and each treatment requires multiple beams. This causes gantries to compete for the beam and results in long beam wait time, which negatively impacts treatment quality and patient satisfaction. In this study, we propose a rollout-based gantry call-back control method, considering both beam wait time and beam utilization. When a gantry becomes available, an optimal call-back delay time is applied to mitigate beam request conflicts. Simulation experiments suggest the proposed method can be used to find the trade-off between beam wait time and the bean utilization and improve proton therapy care delivery process.

Keywords: Medical Robots and Systems, Computer Vision in Automation
Abstract: Malaria causes more than 200 million clinical illnesses and 45 million deaths every year, making mass production of an effective vaccine increasingly urgent. A Plasmodium falciparum sporozoites (PfSPZ) based vaccine has been proved to be a promising choice to defend against the malaria pandemic. However, large scale industrial production of PfSPZ vaccine is currently sub-optimal as the process for sporozoite extraction from salivary glands of infected mosquitoes is performed by manual microdissection, a relatively inefficient process requiring many hours of training. This paper reports continued progress in our development of a robotic system for automating the extraction of salivary glands. Compared to our previous versions, the structure of the new system is optimized to be more compact and allows different steps in workflow to be performed in parallel. We also report continued progress in developing key subsystem components. Finally, experiments show encouraging results with success rates of 90% in robotic mosquito manipulation, 95.6% in salivary gland extraction and 92.7% in debris disposal.

Keywords: Big-Data and Data Mining, AI and Machine Learning in Healthcare, Clinical and Operational Decision Support
Abstract: As one of the essential productive factors, data becomes the fuel driving economic and technological growth. As a result, methods to assess the value of data is necessary in model decision optimization, especially for online platforms which possess a large amount of user data. Prior to this article, effective data valuation algorithms are lack of utilization in online medical platforms. We develop a framework of using an effective data valuation methods based on machine learning, and apply it to the medical data provided by online platform.

Keywords: Optimization and Optimal Control, Industrial and Service Robotics, Planning, Scheduling and Coordination
Abstract: The transportation of long rod-like payload using multiple Multi-Rotor Aerial Vehicles (MRAVs) is analyzed to determine the ideal locations of the anchor points on the payload in order to improve the endurance of the system. The payload is modelled as a beam, with supports as the anchor points of the cable links between the payload and the MRAVs. The criterion for determining the ideal anchor point locations are laid. The proposed method is validated for different types of mass distributions and can be extended to any variation in payload mass distribution transported using multiple MRAVs. The proposed methodology is validated using Hector quadrotor in Gazebo environment.

Keywords: Optimization and Optimal Control, Planning, Scheduling and Coordination, Semiconductor Manufacturing
Abstract: Semiconductor manufacturing includes the most complex manufacturing processes. Scheduling problems to be addressed at the operational level involve a rich set of constraints and criteria. As a result, optimization algorithms are increasingly preferred over dispatching rules, especially in complex production areas such as the photolithography area which is considered in this paper. After a brief problem description, the outlines of a multi-objective scheduling model is presented. A formal study of the relationship between the objective functions is then shown along with numerical experiments.

Keywords: Optimization and Optimal Control, Logistics, Probability and Statistical Methods
Abstract: We develop a continuous asymptotic approximation of the traveling salesman problem with time windows in the Euclidean plane, constructing upon the well-known Beardwood-Halton-Hemmersley theorem. The time windows are taken to be a partition of a given time horizon. Computational experiments on random TSP with time windows instances show that the proposed asymptotic approximations of tour lengths and arrival times are close to the actual optimal values.

Keywords: Optimization and Optimal Control, Planning, Scheduling and Coordination, Intelligent Transportation Systems
Abstract: New mathematical concepts are often only presented from an academic perspective, but this work attempts to highlight the industrial aspects of the implementation of a new concept. In the dense railway system of the Paris area, small delays can easily propagate and disturb the stability of the network, leading eventually to significant perturbations. This situation constitutes a large, real time, multi-objective rescheduling problem. This problem is solved through heuristics and using a variation of Pareto-dominance: Matrix-dominance. After presenting the industrial and academic contexts, this paper discusses possible applications of matrix-dominance and the challenges behind the industrial implementation of this new concept.

Keywords: Optimization and Optimal Control, Calibration and Identification, Model Learning for Control
Abstract: The recent trends of the COVID-19 research are being devoted to disease transmission modeling in presence of vaccinated individuals, while the emerging needs are being focused on developing effective strategies for the optimal distribution of vaccine between population. In this context, we propose a novel non-linear time-varying model that effectively supports policy-makers in predicting and analyzing the dynamics of COVID-19 when partially and fully immune individuals are included in the population. Differently from the related literature, where the common strategies typically rely on the prioritization of the different classes of individuals, we propose a novel Model Predictive Control approach to optimally control the multi-dose vaccine administration in the case the available number of doses is not sufficient to cover the whole population. Focusing on the minimization of the expected number of deaths, the approach discriminates between the number of first and second doses. We calibrate the model on the Israeli scenario using real data and we estimate the impact of the vaccine administration on the virus dynamics. Lastly, we assess the impact of the first dose of the Pfizer's vaccine confirming the results of clinical tests.

Keywords: Hybrid Logical/Dynamical Planning and Verification, Optimization and Optimal Control
Abstract: This paper proposes an approach to the problem of capacity planning in hospitals during the COVID-19 crisis. It proposes three types of methods: dynamic, static, and hybrid. We have implemented them on several epidemic patterns and for two distinct problems. These problems are distinguished by the possibility or not of accepting patients in conditions of overcrowding.

Keywords: Inventory Management, Logistics, Manufacturing, Maintenance and Supply Chains
Abstract: To efficiently meet demand in a production system, the lot-sizing problem determines a production plan that minimizes the overall costs, optimizes the use of the available resources, and satisfies demand requirements. Nonetheless, uncertainties in the production environment directly affect the quality and feasibility of the production plans. In fact, demand can be highly volatile and influenced by multiple factors such as age, life-cycle, economic context, reference groups, culture, festive season. To increase the robustness of the production plan to unforeseen uncertainties, one could rely on the robust optimization methodology that offers ease and flexibility to account for uncertain parameters. In the light of the robust approaches, an adaptive robust uncapacitated lot-sizing model is proposed to deal with an uncertain demand. It offers a production plan that can be updated when demand information unfolds over time. Numerical experiments demonstrate that the adaptive model outperforms the static model, while a marginal additional computational effort is required to obtain a robust production plan. The results also indicate that the proposed approach is a better alternative for production planning within a system that is flexible for changes in the lot size at each period.

Keywords: Manufacturing, Maintenance and Supply Chains
Abstract: This paper deals with a comparison of production control policies in a two-product two-echelon supply chain dynamic problem with production capacity constraint. The factory related echelon consists of an unreliable manufacturing system that cannot produce both types of products simultaneously. Therefore, changeover operations are required to switch from one type of product to another. The decision on the product changeover depends on the production control policy. This research compares the well-known Hedging Corridor Policy, which has been previously adopted by the literature in a supply chain with production capacity constraints, and the Improved Modified Hedging Corridor Policy, which has been proved to minimize the total cost incurred in manufacturing companies characterized by unreliable production systems with constant demand rate. For comparison purposes, an experimental campaign was conducted through an analytical model based on discrete-time difference equations to investigate the Fill Rate of the multi-product supply chain as response variable. A proper ANOVA analysis and a set of interval plots revealed that the Hedging Corridor Policy outperforms the Improved Modified Hedging Corridor Policy in improving the Fill Rate Indicator.

Keywords: Manufacturing, Maintenance and Supply Chains, Planning, Scheduling and Coordination, Intelligent and Flexible Manufacturing
Abstract: In supply chains the collaboration between manufacturing companies is crucial to ensure resilient and efficient operation in the face of uncertainty. This is especially true for manufacturers with global supply chains in a highly competitive market with fluctuating demand. We develop a novel approach to sharing of manufacturing capacity in supply chains. The approach is based on combinatorial auctions in which suppliers and buyers submit bids for manufacturing capacity. We present the mixed integer programming formulation of the winner determination problem and evaluates the efficiency and complexity of the approach on several realistic instances. We find that the number of accepted bids increases up to 10-fold compared to a situation without capacity sharing.

Keywords: Planning, Scheduling and Coordination, Manufacturing, Maintenance and Supply Chains, Logistics
Abstract: The operational management of crude oil supply entails solving large-scale mixed-integer nonlinear programming (MINLP) problems, accounting for unloading and transfer operations in terminals, inventory control, and blending of crude oils to meet the demands from the refinery. In offshore oil assets, the planning of operations becomes more challenging because vessels' trips should be scheduled to relieve production platforms from crudes transferred to the terminals. Arguably, the problem's computational hardness emerges from its size and the combination of discrete decisions with nonlinear constraints, consisting of bilinear terms that model blending operations. Concerning the nonlinear functions, this work contributes by developing a linear approximation of the bilinear terms with the multiparametric disaggregation technique (MDT). The MDT application yields a mixed-integer linear programming (MILP) relaxation, which can be combined with a local nonlinear programming (NLP) algorithm to reach a feasible schedule of operations.

Keywords: Planning, Scheduling and Coordination, Manufacturing, Maintenance and Supply Chains, Inventory Management
Abstract: This paper first motivates the need to consider the dynamic evolution of the inventory within periods when solving the capacitated lot-sizing problem with lost sales, setup times and inventory bounds. Two models are introduced to characterize and bound the minimum and maximum inventory levels within each period under specific assumptions on the production and the demand shapes. Computational experiments are carried out to compare these models and emphasize the need for a better inventory modeling in lot-sizing problems.

Keywords: Sustainable Production and Service Automation, Energy and Environment-Aware Automation
Abstract: Energy efficiency is becoming a key subject in manufacturing, especially if related to the major environmental impact of machining activities. Nowadays, great research efforts are performed to find new methodologies to improve sustainability of manufacturing processes. Energy consumption can be lowered controlling machine state during idle periods. This can be achieved with Energy-Efficient Control (EEC) policies that switch off/on the machine. The same approach can be applied simultaneously to more identical devices in a parallel machines workstation. This paper presents a novel model to identify EEC policies for manufacturing systems composed of parallel and identical machines with finite buffer capacity. In this configuration, each machine can be switched on with a stochastic startup time and switched off instantaneously also considering an availability constraint. The proposed model reduces energy consumption while assuring a target availability level. The control is executed using buffer level information. Numerical results confirm model benefits when applied to a real industrial system from the automotive sector.

Keywords: Automation in Life Science: Biotechnology, Pharmaceutical and Health Care, Process Control, Robust/Adaptive Control
Abstract: An adaptive control algorithm for set-point tracking of bioreactor-scale microbial cultivation process parameters is presented, where adaptation of PI controller parameters is based on the controller input/output signals only and does not require additional measurements of process variables. A simple gain scheduling algorithm is developed using tendency model of the controlled process. Extreme operating conditions were selected and modeled to test the algorithms performance while controlling pH in fed-batch operating mode. The simulation results showed that the gain-scheduled controller demonstrated robust behavior and significantly improved the control performance compared to that of fixed-parameter controllers.

Keywords: Foundations of Automation
Abstract: In this paper, the Lasso formulation of the Wahba's problem is studied for spacecraft attitude determination problem from vector observations. This problem is designed in order to make the attitude solution more regularized and robust. The analytical quaternion solution is derived and is proven to be more continuous than results from previous Wahba's solutions. First, we point out its application in quaternion conversion from rotation matrix. Then, simulation tests are achieved using real motion and simulated three vector observation pairs from a star tracker, a sun sensor and a magnetometer. The results well support the proposed method for attitude determination and verify its effectiveness.

Keywords: Robust/Adaptive Control, Autonomous Vehicle Navigation, Agricultural Automation
Abstract: In this work, we present a new robust vision-based controller for wheeled mobile robots, equipped with a fixed monocular camera, to perform autonomous navigation in agricultural fields accurately. Here, we consider the existence of uncertainties in the parameters of the robot-camera system and external disturbances caused by high driving velocities, sparse plants, and terrain unevenness. Then, we design a robust image-based visual servoing (rIBVS) approach based on the sliding mode control (SMC) method for robot motion stabilization, even under the presence of such inaccuracies and perturbations. The vision-based controller, based on column and row primitives, is slightly modified to include a robustness term into the original feedback control laws to ensure successful row crop reaching and following tasks. We employ the Lyapunov stability theory to verify the stability and robustness properties of the overall closed-loop system. 3D computer simulations are carried out in the ROS-Gazebo platform, an open-source robotics simulator, using a differential-drive mobile robot (DDMR) in an ad-hoc developed row crop environment to illustrate the effectiveness and feasibility of the proposed control methodology.

Keywords: Foundations of Automation, Intelligent Transportation Systems, Motion and Path Planning
Abstract: In this paper a multi-destination multi-source traffic network of a single kind of vehicles is considered. In particular, an analysis of the system model in steady state conditions has been performed both as regards the input flows from the external and the splitting rates. It has been investigated under which conditions the system admits a unique asymptotically stable equilibrium point. The condition that will be described has a very clear physical significance and it represents an interesting insight about the way by which the unique asymptotically stable equilibrium point can be found analytically when the information about the external inputs, the values of the splitting rates, and the parameters characterizing each link in the network are made available.

Keywords: Robust/Adaptive Control, Motion Control, Prosthetics and Exoskeletons
Abstract: Nowadays, legged mobile robots have increased the interest of the robotics community because such mechanisms have higher versatility and autonomy compared to wheeled mobile robots. Although single-leg or multi-leg mechanisms can cross any terrain, some disadvantages are related to their increased complexity in mechanical design, modeling and control, and higher power consumption. A first case study for the balance and motion planning problem is the hopping robot, which is a nonholonomic system whose motion dynamics of each hopping cycle can be split into flight and stance phases. In this work, we consider the modeling and control design of a one-legged hopping robot in the flight phase by using the sliding mode approach, due to its well-known ability to deal with parametric uncertainties and nonlinear disturbances. Then, two nonlinear controllers are designed and implemented to automatically stabilize the robot joints during the flight in the presence of perturbations caused by the neglected higher-order nonlinear terms in the modeling process, unmodeled dynamics and measurement noise. The Lyapunov stability theory is used to demonstrate the stability and robustness properties of the overall closed-loop control systems. Numerical simulations and a comparative analysis are provided to illustrate the performance and feasibility of the proposed control methodology.

Keywords: Foundations of Automation, Industrial Robots, Intelligent and Flexible Manufacturing
Abstract: This work presents an iterative learning control (ILC) algorithm to enhance the feedforward control (FFC) for robotic manipulators. The proposed ILC algorithm enables the cooperation between the ILC, inverse dynamics, and a PD feedback control (FBC) module. The entire control scheme is elaborated to guarantee the control accuracy of the first implementation; to improve the control performance of the manipulator progressively with successive iterations; and to compensate both repetitive and non-repetitive disturbances, as well as various uncertainties. The convergence of the proposed ILC algorithm is analysed using a well established Lyapunovlike composite energy function (CEF). A trajectory tracking test is carried out by a seven-degree-of-freedom (7-DoF) robotic manipulator to demonstrate the effectiveness and efficiency of the proposed control scheme. By implementing the ILC algorithm, the maximum tracking error and its percentage respect to the motion range are improved from 5:78 to 1:09degree,and 21:09% to 3:99%, respectively, within three iterations.

Keywords: Logistics, Health Care Management, Planning, Scheduling and Coordination
Abstract: This paper introduces a transition flow model to study fall-related emergency department (ED) revisits for elderly patients with diabetes. Five diabetes classes are used to classify patients at discharge, within 7-day revisits, and between 8 and 30-day revisits. Analytical formulas to evaluate patient revisiting risks are derived. To reduce revisits, sensitivity analysis is introduced to identify the most critical, i.e., dominant, factors whose changes can lead to the largest reduction in revisits. In addition, a case study at University of Wisconsin (UW) Hospital ED is described to illustrate the applicability of the model.

Keywords: Health Care Management, Robotics and Automation in Life Sciences, Building Automation
Abstract: Sit-to-stand (STS) movement is a frequently performed activity of daily living. The analysis of STS movement can lead to the early detection of a decline in physical functioning. There is thus interest in home-based STS monitoring systems. The present study (1) extracted universal differences by age group in common STS movements of daily life and (2) clarified where the differences between age groups lie in the sequence of movements. We conducted an experiment to measure the three patterns of STS movement assumed to be performed in living spaces. Ten young adults and thirteen elderly people participated. Data were acquired by an Azure Kinect DK, an RGB-D camera, and converted into angles and angular velocities of the major body segments. Principal component analysis was applied to the overall time series data. From the synthesized principal component space, principal components with significant differences between the elderly and young-adult groups were extracted, and movements of the elderly participants were evaluated. We found parameters and movement strategies that show the characteristics of each age group for each movement pattern. The results of this study can contribute to the design of home-based STS monitoring systems.

Keywords: AI and Machine Learning in Healthcare, Probability and Statistical Methods
Abstract: Medical test selection is a recurring problem in health prevention and consists of proposing a set of tests to each subject for diagnosis and treatment of pathologies. The problem is characterized by the unknown risk probability distribution across the population and two contradictory objectives: minimizing the number of tests and giving the medical test to all at-risk populations. This article sets this problem in a general framework of chance-constrained medical test rationing with unknown subject distribution over an attribute space and unknown risk probability but with a given sample population. A new approach combining decision-tree and Bayesian inference is proposed to allocate relevant medical tests according to the subjects’ profile. Case studies on screening of hypertension and diabetes are conducted, and the performance of the proposed approach is evaluated. Significant savings on unnecessary tests are achieved with limited numbers of subjects needing but not receiving necessary tests.

Keywords: Health Care Management, Sensor Networks
Abstract: In the context of the Internet of Things (IoT) healthcare, biophysical features collected during sleep needs robust analysis methods to be efficiently used to detect sleep disorders. In this paper, analysis methods using a limited number of input variables (cardiac, respiratory, and body movement) have been used to perform the classification of sleep stages. The efficiency of each classification method has been compared to a reference method that combines a large number of biophysical features referred to as PolySomnoGraphy (PSG). Five classical machine learning methods were evaluated by testing their accuracy on the same collected data. Finally, using a neural network with a short memory method, the classification task fitted 91.34% of the PSG classification.

Keywords: Health Care Management, Inventory Management
Abstract: Hospitals worldwide share common challenges in finding the optimal number of patient beds in advance, which is referred to as the bed capacity management problem. Effective management cannot be achieved without an appropriate understanding of the stochastic nature of patient demands. In light of the widespread usage of newsvendor models in inventory management and the increasingly available operational data, we propose a data-driven approach to tackling the supply-demand matching problem under uncertainty, factoring in the observation that daily hospitalization demands are not independent and identically distributed. We develop forecast-based newsvendor models that use autoregressive integrated moving average (ARIMA) to make predictions, and determine the optimal bed capacity during the designated forecast periods. The performance of this approach is tested using real-world data from the West China Hospital. Our analysis affirms the superiority of the forecast-based models in practical settings and further enlightens the situations where forecast might not be effective --- when the normality assumption of the noise distribution does not strictly hold, the empirical distribution model can be a computationally cost-effective choice.

Keywords: Computer Vision for Transportation, Intelligent and Flexible Manufacturing, Logistics
Abstract: Cuboid objects are widely used in logistics and manufacturing, since they are ideal for automatic manipulation. For that, a reliable detection, identification and tracking algorithm of cuboid objects is a key requirement. RGBD cameras are suited for these tasks, however they suffer from limitations such as occlusions from clutter or a small field of view. A multi RGBD camera setup would remedy many of these limitations. In this work, we present a cuboid detection, identification and tracking system using a multi RGBD camera setup. First, we calibrate our multi camera setup using a robot arm and a colored spherical calibration target. By using spatial and color information, a preprocessing pipeline segments then the unified point cloud into locally connected planar segments. Afterwards, custom spatial criteria allows for assigning the resulting segments to cuboid groups. We present a new modified bounding box regression method to estimate cuboid boxes and a novel intersection over union based approach to track cuboids across frames. By applying our approach, the automatic manipulation and transportation in manufacturing industry will become more efficient, which helps to save more time and cost. Our system is tested and analyzed with a recorded dataset, which contents variant constellations. The reliability and accuracy are demonstrated.

Keywords: Computer Vision for Transportation, Intelligent Transportation Systems, Computer Vision in Automation
Abstract: Among several components of post-accident procedures, the automation of car part recognition on images benefits efficient operations of car sharing platforms and insurance companies. The prior studies employed various computer vision algorithms to automate car part recognition tasks under the supervised learning paradigm; however, these studies bear several drawbacks to be applied in the real world. The supervised approaches required manual annotations on the dataset, and they assumed car part images always include the global shape of the vehicle while the real-world images do not follow the assumption. In pursuit of improving these limits, our study proposed a novel approach to automating car part recognition with the following contributions. First, we examined the self-supervised feature extractor better understands the visual representation of car part images rather than conventional methods. Second, we scrutinized resizing with interpolation and augmenting with the normalization most effectively highlights car part images' visual patterns. Third, we designed an automated car part recognition system with fewer human interventions than prior research. Lastly, we examined our automation approach trained with midsize car images can be transferred into other car types; thus, the practitioner can save resources in real-world practices. Although our automated car part recognition approach still bears a human intervention, we expect it would be a concrete baseline of further automation approaches to accomplish efficient post-accident procedures in the real world.

Keywords: Computer Vision in Automation, Autonomous Vehicle Navigation, Agricultural Automation
Abstract: Local perception navigation methods allow agricultural robots to accurately track crop row structures while performing automated farming tasks. The integration of these methods as a part of a fully autonomous navigation solution requires continuous assessment of their reliability since they rely solely on sensor data in a changing and unpredictable environment. This paper presents a data-driven monitoring approach for the task of structure-based navigation in agriculture. The proposed method applies semi-supervised anomaly detection, aiming to learn a model of normal scene geometry that characterizes a domain of reliable execution of the considered task. To this end, a convolutional neural network was trained in one-class classification fashion on Hough representations of LiDAR point clouds. In experimentation, the learned normal model was used to derive a confidence measure for a LiDAR-based tracking algorithm allowing its integration as a part of a hybrid navigation solution in vineyards for a commercial robotic platform.

Keywords: Computer Vision for Manufacturing, Deep Learning in Robotics and Automation, Machine learning
Abstract: Steel is one of the most important building materials of our time, and the process of producing flat plates is complex. Before steel is shipped or delivered, the sheets must undergo a thorough inspection procedure to avoid defects. Identification and classification of rolled metal surface defects are one of the main tasks for the correct evaluation of product quality. This work aims to develop a method for recognizing and classifying defects of metal surfaces by their images in real-time. The algorithm is aimed at improving production standards and process efficiency. In this paper, Deep Learning (DL) and Computer Vision (CV) techniques are used to solve the problem of defect detection on the surface of steel sheets. Convolutional Neural Network (CNN) architectures are compared, and various steel defects are detected and recognized. The result of this work is a comparative analysis of DL models and the choice of an algorithm designed to search and classify defects in real-time. The use of one CNN model can make it possible to create a tool that greatly facilitates the work of a person.

Keywords: Computer Vision for Manufacturing, Logistics, Computer Vision in Automation
Abstract: In industrial part kitting, 3D objects are inserted into cavities for transportation or subsequent assembly. Kitting is a critical step as it can decrease downstream processing and handling times and enable lower storage and shipping costs. We present Kit-Net, a framework for kitting previously unseen 3D objects into cavities given depth images of both the target cavity and an object held by a gripper in an unknown initial orientation. Kit-Net uses self-supervised deep learning and data-augmentation to train a convolutional neural network (CNN) to robustly estimate 3D rotations between objects and matching concave or convex cavities using a large training dataset of simulated depth images pairs. Kit-Net then uses the trained CNN to implement a controller to orient and position novel objects for insertion into novel prismatic and conformal 3D cavities. Experiments in simulation suggest that Kit-Net can orient objects to have a 98.9 % average intersection volume between the object mesh and that of the target cavity. Physical experiments with industrial objects succeed in 18 % of trials using a baseline method and in 63 % of trials with Kit-Net. Video, code, and data are available at https://github.com/BerkeleyAutomation/Kit-Net

Keywords: Computer Vision for Transportation, Intelligent Transportation Systems, Computer Vision for Automation
Abstract: Taking advantage of the complementary error characteristics of Light Detection and Ranging (LiDAR) and stereo camera reconstruction, we propose a set-membership-based method for fusing LiDAR information with dense stereo data under consideration of interval uncertainty of all measurements and calibration parameters. Employing interval analysis, we can propagate the uncertainties to the extraction of distinct features in a straightforward manner. To show the applicability of our approach, we use the fused information for dead reckoning. In contrast to other works, we can consistently propagate the sensor uncertainties to the localization of the robot. Further, we can provide guaranteed bounds for the relative motion between consecutive frames. Using real data we validate that our approach is indeed able to always enclose the true pose of the robot.

Keywords: Motion and Path Planning, Reactive and Sensor-Based Planning, Industrial and Service Robotics
Abstract: Mechanical search, the finding and extracting of a known target object from a cluttered environment, is a key challenge in automating warehouse, home, retail, and industrial tasks. In this paper, we consider contexts in which occluding objects are to remain untouched, thus minimizing disruptions and avoiding toppling. We assume a 6-DOF robot with an RGBD camera and unicontact suction gripper mounted on its wrist. With this setup, the robot can move both camera and gripper in order to identify a suitable approach vector, reach in to achieve a suction grasp of the target object, and extract it. We present AVPLUG: Approach Vector PLanning for Unicontact Grasping, an algorithm that uses an octree occupancy model and Minkowski sum computation to find a collision-free grasp approach vector. Experiments in simulation and with a physical Fetch robot suggest that AVPLUG finds an approach vector up to 20 times faster than a baseline search policy.

Keywords: Motion and Path Planning, Autonomous Agents, Collision Avoidance
Abstract: The paper focuses on the problem of 3D path planning for the underwater vehicle in an environment with obstacles. An efficient framework is presented where environment modelling and path planning are effectively performedto obtain a optimized feasible path. Firstly, octree is adopted to divide the spatial environment into grids non-uniformly. Hierarchical distances between adjacent grids in the tree are deduced and are then used to determine grid adjacent relationship. The environment modelling method would be helpfulnot only to reduce memory consumption but also to improve computational efficiency while maintaining modeling accuracy.Consequently, an improved heuristic search strategy is proposed by considering the features of underwater vehicle to obtainmore feasible intermediate path points. Once all intermediate points are determined, they are globally optimized with a correction strategy and then, a cubic bezier curve is adopted to smooth the points. The resulting path satisfies nonholonomic constraints and makes the vehicle avoid obstacles autonomously.The proposed approach is tested on simulation and realisticunderwater terrain scenes, proven to be efficient to generate feasible 3D trajectories with high real-time performance. The approach has important application value in the guidance ofunderwater vehicle.

Keywords: Motion and Path Planning, Optimization and Optimal Control, Collaborative Robots in Manufacturing
Abstract: The purpose of this paper is to propose an energy-efficient pick-and-place motion planning that a robot grabs a workpiece on a belt-conveyor and releases it to specified position for saving energy of a dual-arm robot. The motion is evaluated by the objective function that has the number overshooting, the time of overshooting, the time of termination and the consumption of energy for the total minimization problem. The robot arm is controlled by PID controller, and its optimal PDI gains are obtained by a Genetic Algorithm. We conducted the numerical simulation to verify that the consumption of energy using a dual-arm robot can be decreased compared with that of one arm. The proposed motion planning has been implemented into real world motion planning problem. A dual-armed robot named duAro released by Kawasaki Heavy Industries, is used to carry out the real implementations to test the proposed pick-and-place motion planning. The experimental results show the effectiveness of the dual-armed motions compared with the single-arm motions for energy efficiency.

Keywords: Motion and Path Planning, Model Learning for Control, Planning, Scheduling and Coordination
Abstract: Teaching a robot to learn calligraphy writing has been an interesting and challenging topic for robot learning. Ideally, to achieve human-like behavior, a robot can imitate any character fonts out of image inputs only. In the past decades, many works have been done in different kinds of special calligraphy robots design and learning by demonstration. Recently, advanced learning algorithms such as reinforcement learning have also been used with a significant sacrifice in data collection and computational complexity. In this paper, we present a simple and compact hybrid learning approach, by combining offline learning in simulator and online motion planing. In our approach, we simplify the writing trajectory generation to an optimization problem considering the width of the stroke as a function of height (z-axis) only. Based on the skeleton from each extracted stroke, Dynamic programming and Gaussian process models are used to solve the widths at each sampling point and to convert into a smooth trajectory. In such manner, a robot can learn to write any Chinese character directly from an image input within half a minute.

Keywords: Motion and Path Planning, Optimization and Optimal Control, Industrial and Service Robotics
Abstract: Motion planning for kinematic redundancy is an area of great importance for maximizing the mobility of robotic systems. However, generating optimized motions for this type of system is a challenging task given the large search space of possible configurations. Previously proposed methods do not address path following tasks with constrained end-effector position and orientation for a mobile manipulator system with more than 6 degrees of freedom (DoF). This paper presents a novel computational method for simultaneous optimization of base and manipulator robotic system with 8 DoF for welding tasks, constraining both end-effector position and orientation. The mobile manipulator consists of a 2 DoF non-holonomic base and a 6 DoF manipulator. The proposed method applies a Genetic Algorithm (GA) to solve for optimized configurations for the base and manipulator for strategically sampled end-effector waypoints. The base configurations and end-effector orientations are interpolated between the GA solutions, and used as inputs for an inverse kinematics solver to find the optimal manipulator pose. The experiment results show that the proposed methods create optimized smooth and continuous motions for both the base and manipulator, while constraining the end-effector position and orientation. The proposed method is a novel application of GA optimization, with improved results for path following motion planning by including sampling, interpolation, and inverse kinematics steps within the methodology.

Keywords: Motion and Path Planning, Planning, Scheduling and Coordination, Sensor Networks
Abstract: Persistent monitoring on terrains using mobile robotic sensors requires coordinated planning. Terrain features add visibility obstacles and limited fuel capacity of aerial robots leads to range restrictions that make the problem challenging. We address the visual-monitoring problem on piecewise linear features within a terrain using multiple mobile robots for persistent operations. The planner must account for visual coverage, refueling aerial robots during the mission, and placement of refueling depots while also utilizing the available sensor diversity to minimize overall costs for the monitoring mission. Building on previous works on visibility in specific classes of polygons and fuel-constrained routing, we develop a discrete representation of the problem that allows the design and application of discrete optimization techniques to find optimal solutions. We develop a mixed-integer linear programming (MILP) formulation and discuss a branch-and-cut implementation to compute exact solutions. We also develop a construction heuristic based on the idea of competitive construction of robot paths using a step-increment strategy. We report the results from computational simulations and illustrate proof of concept using experiments on real robots

Keywords: Model Learning for Control, Control Architectures and Programming, Motion Control
Abstract: A hybrid filtered basis function (FBF) approach is proposed in this paper for feedforward tracking control of linear systems with unmodeled nonlinear dynamics. Unlike most available tracking control techniques, the FBF approach is very versatile; it is applicable to any type of linear system, regardless of its underlying dynamics. The FBF approach expresses the control input to a system as a linear combination of basis functions with unknown coefficients. The basis functions are forward filtered through a linear model of the system’s dynamics and the unknown coefficients are selected such that tracking error is minimized. The linear models used in existing implementations of the FBF approach are typically physics-based representations of the linear dynamics of a system. The proposed hybrid FBF approach expands the application of the FBF approach to systems with unmodeled nonlinearities by learning from data. A hybrid model is formulated by combining a physics-based model of the system’s linear dynamics with a data-driven linear model that approximates the unmodeled nonlinear dynamics. The hybrid model is used online in receding horizon to compute optimal control commands that minimize tracking errors. The proposed hybrid FBF approach is shown in simulations on a model of a vibration-prone 3D printer to improve tracking accuracy by up to 65.4%, compared to an existing FBF approach that does not incorporate data.

Keywords: Model Learning for Control, Deep Learning in Robotics and Automation, Telerobotics and Teleoperation
Abstract: Recent insights from human-robot intelligence and deep learning raise hope towards task-specific autonomy in robotic intravascular coronary interventions. However, lack of learning-based methods for characterizing the interventionists’ kinesthetic data hinders the drive for shared control and robotic autonomy during cyborg catheterization. In this study, a deep multimodal network model is proposed for classification and recognition of interventionists’ hand movements during cyborg intravascular catheterization. The model has two modules for extracting salient features in electromyography signal datasets, and classification of hand motions made during intravascular catheterization procedures. Network training and evaluation observed for in-vitro and in-vivo datasets obtained from trained novice subjects and expert with about 5 years of experience in percutaneous coronary interventions. Performance evaluation shows the learning model could classify interventionists’ hand movements accurately in manual and robot-assisted navigations, respectively. This study is suggested to further stimulate the development of appropriate skill level assessments towards cyborg catheterization for cardiac interventions.

Keywords: Model Learning for Control, Robust/Adaptive Control, Learning and Adaptive Systems
Abstract: Many robot manipulation skills can be represented with deterministic characteristics and there exist efficient techniques for learning parameterized motor plans for those skills. However, one of the active research challenge still remains to sustain manipulation capabilities in situation of a mechanical failure. Ideally, like biological creatures, a robotic agent should be able to reconfigure its control policy by adapting to dynamic adversaries. In this paper, we propose a method that allows an agent to survive in a situation of mechanical loss, and adaptively learn manipulation with compromised degrees of freedom--- we call our method Survivable Robotic Learning (SRL). Our key idea is to leverage Bayesian policy gradient by encoding knowledge bias in posterior estimation, which in turn alleviates future policy search explorations, in terms of sample efficiency and when compared to random exploration based policy search methods. SRL represents policy priors as Gaussian process, which allows tractable computation of approximate posterior (when true gradient is intractable), by incorporating guided bias as proxy from prior replays. We evaluate our proposed method against off-the-shelf model free learning algorithm (DDPG), testing on a hexapod robot platform which encounters incremental failure emulation, and our experiments show that our method improves largely in terms of sample requirement and quantitative success ratio in all failure modes. A demonstration video of our experiments can be viewed at: https://sites.google.com/view/survivalrl

Keywords: Model Learning for Control
Abstract: Interaction dynamics are difficult to model analytically, making data-driven controllers preferable for contact-rich manipulation tasks. In this work, we approximate the intricate dynamics of food-cutting with a Long Short-Term Memory (LSTM) model to apply a Model Predictive Controller (MPC). We propose a problem formulation that allows velocity-controlled robots to learn the interaction dynamics and tackle the difficulty of multi-step predictions by training the model with a horizon curriculum. We experimentally demonstrate that our approach leads to good predictive performance that scales for longer prediction horizons, generalizes to unseen object classes and results in controller behaviors with an understanding of the cutting dynamics.

Keywords: Remanufacturing, Optimization and Optimal Control
Abstract: Remanufacturing is an efficient approach to recycle end-of-life (EOL) product, where disassembly is a critical step. Disassembly sequence planning problems aim at determining a sequence of disassembly operations to optimize some specified criteria. This work considers multiple EOL products that need to be disassembled in a planning horizon. It proposes a stochastic profit-oriented multi-product disassembly sequence planning problem for the first time, where some uncertainties during a disassembly process are considered. To solve it, an enhanced equilibrium optimizer integrated with a stochastic simulation approach is developed. A double-link string is applied to represent a solution. A decoding approach is developed to convert it to a feasible solution. Then, a local-best search approach is designed to search the promising regions. Besides, a stochastic simulation method is adopted to measure the feasibility and performance of the obtained solutions. Through experiments on some real cases and comparisons with three competitive approaches, the effectiveness of the developed method in addressing the proposed problem is validated

Keywords: Intelligent and Flexible Manufacturing
Abstract: The work aims to solve a cell formation and product scheduling problem by considering dual-resource setting, supply chain, profit maximization, and operational error rate in multi-factory cellular manufacturing systems. This problem is important and highly challenging. To solve it, this work proposes a genetic algorithm with novel initialization. Computational experiments are conducted to show that the proposed algorithm has better performance than conventional genetic algorithm without the proposed heuristic given the same computational budget.

Keywords: Diagnosis and Prognostics, Failure Detection and Recovery, Sensor Fusion
Abstract: The health condition of industrial equipment is closely related to productivity and safety, attaching great importance to fault diagnosis. Although current fault diagnosis methods have already achieved good effect to some degree, it is more reliable and outstanding to utilize multiple signal sources and capture attention to the effective information differences between various signals. In order to take full advantage of fault information, this paper proposes a new method of fault diagnosis based on multi-source signals and attention mechanism. This method learns the weights of multi-source features with attention mechanism and recalibrates the feature responses. Besides it extracts and fuses the fault features based on residual network (ResNet). The experimental results show that proposed method can effectively learn the differences of multi-source signals, and has excellent classification performance compared to those based on a single signal or multiple signals without attention mechanism.

Keywords: Semiconductor Manufacturing, Petri Nets for Automation Control, Planning, Scheduling and Coordination
Abstract: We address a scheduling problem of dual-gripper robotic cells in which multiple different jobs with the same process flow are processed with the objective of makespan. We propose a novel approach that provides an optimal solution by using state ready times of a Petri net. The experimental results show the efficiency of the proposed algorithm.

Keywords: Intelligent and Flexible Manufacturing
Abstract: To ensure stable manufacturing and high yield of production, factories (e.g., semiconductor or TFT-LCD fabs) conduct quality inspection on workpieces. They tend to adopt sampling inspection in consideration of reducing cost and cycle time, yet that fails to achieve real-time and online total inspection because of the sampling strategy and metrology delay. Automatic Virtual Metrology (AVM) is the best solution to tackle the problem mentioned above, due to the fact that it can convert sampling inspection with metrology delay into on-line and real-time total inspection. However, with the advancement of science and technology, the processes become more and more sophisticated, and the requirement for the accuracy of virtual metrology becomes higher. The current AVM prediction algorithm is the traditional machine learning method, Back-Propagation Neural Networks (BPNN). However, even if the amount of data in this method increases, the performance improvement has its limits, and it requires a strict and time-consuming feature selection process. To improve the prediction accuracy, this work proposes the deep learning method, Convolutional Neural Networks (CNN), for the AVM server. The accuracy of CNN improves as the amount of data grows. In other words, if there are sufficient data, the current accuracy limit of machine learning can be enhanced. Experimental results reveal that CNN can automatically extract highly informative features from the data and improves the original AVM accuracy.

Keywords: Planning, Scheduling and Coordination, Optimization and Optimal Control, Simulation and Animation
Abstract: The economy of scale concept has clearly shown the connection between the change in production quantity and the associated price development over the past few decades. In the following, a formula is presented, roughly calculating future costs based on volume changes and constant cost reduction per doubling. This formula can find application in the field of supply chain planning and supply chain simulations, and thus could lead to data-driven automation.

Keywords: Task Planning, Deep Learning in Robotics and Automation, AI-Based Methods
Abstract: In recent years, the robotics community has made substantial progress in robotic manipulation using deep reinforcement learning (RL). Effectively learning of long-horizon tasks remains a challenging topic. Typical RL-based methods approximate long-horizon tasks as Markov decision processes and only consider current observation (images or other sensor information) as input state. However, such approximation ignores the fact that skill-sequence also plays a crucial role in long-horizon tasks. In this paper, we take both the observation and skill sequences into account and propose a skill-sequence-dependent hierarchical policy for solving a typical long-horizon task. The proposed policy consists of a high-level skill policy (utilizing skill sequences) and a low-level parameter policy (responding to observation) with corresponding training methods, which makes the learning much more sample-efficient. Experiments in simulation demonstrate that our approach successfully solves a long-horizon task and is significantly faster than Proximal Policy Optimization (PPO) and the task schema methods.

Keywords: Task Planning, Optimization and Optimal Control
Abstract: With the rapid growth of social electricity demand, in order to prevent accidental failure of the equipment in the substation, the human resources and operation costs of substation operation and maintenance continue to increase, so the optimization of substation operation and maintenance has attracted great attention. This paper focuses on the optimal coordination of the human resources and the operation and maintenance tasks in the substation. The optimal coordination problem is formulated as a mixed integer linear programming problem. The objective is to minimize the total traffic distance of staff during the execution of tasks which is a direct factor reflecting the cost of operation and maintenance in practice. In the optimization model developed in this paper, we formulate the task execution time in a continuous way, which can avoid the trade-off between calculation accuracy and efficiency, even the infeasibility, caused by time discretization. Furthermore, each task of the substation operation and maintenance is defined as an event for the problem. In this way, the objective function is independent of time and the problem can be efficiently solved based on the triggering of the events. The performance and effectiveness of the developed method is demonstrated using a practical case study.

Keywords: Swarms, Motion and Path Planning, Autonomous Agents
Abstract: Large scale planetary surface exploration by a multi-robot system requires collective behavior and information exchange between the agents, which demands for a valid communication model. This paper aims to validate a proposed deterministic radio propagation model developed towards coordinated path planning of multi-robot systems, by the means of terrestrial experiments. The model developed for planetary surface exploration considers the specific characteristics of the site and low-heighted antennas of micro-rovers to determine the point-to-point signal quality. For validation purposes, the terrain geometry of the experiment site on Earth is studied, and the propagation parameters are adjusted accordingly. The point-to-point communication links for three different carrier frequencies and two different transmitter antenna heights are measured. The model was then updated such that the received signal strength and the path gains from the measured data fit well with the predicted data. Therefore, the developed radio propagation model with the required adjustments can be applied for communication-aware path planning of moving rovers on any planetary surface if the topography of the operating region is known in advance.

Keywords: Swarms, Formal Methods in Robotics and Automation, Task Planning
Abstract: This paper presents a distributed strategy for automatically synthesizing controls for robotic swarms such that they achieve reactive, high-level formation tasks. In our framework, a user specifies formation and location-based swarm tasks, which may include reactions to environmental events, using linear temporal logic. Then, we synthesize a centralized finite automaton that represents the symbolic behavior of the swarm. To execute the automaton, we develop an auction-based decentralized algorithm that assigns robots to different locations and formations using only information from neighboring robots. To guarantee that the swarm can achieve the specified high-level tasks, we use integer programming to obtain the maximum and minimum number of robots that need to be sent to different locations during each symbolic transition, and we incorporate the constraints on sub-swarm sizes into the auction-based assignment algorithm. We demonstrate our control framework in simulation and with ten Anki-Vector robots in the lab.

Keywords: Swarms, Optimization and Optimal Control, Planning, Scheduling and Coordination
Abstract: This paper presents a model predictive control (MPC) based algorithm for tracking multiple targets using a swarm of unmanned aerial vehicles (UAVs). All the UAVs belong to fixed-wing category with constraints on flight velocity, climb rate and turn rate. Each UAV carries a camera to detect and track the target. Two cases are considered where for the first case, the number of the UAVs is equal to the number of targets. For the second case, the number of UAVs is lesser than the number of targets leading to a conservative solution where the objective is to maximize the average time duration for which the targets are in the field-of-view (FOV) of any one of the UAV's camera. A data driven Gaussian process (GP) based model is developed to relate the hyperparameters used in MPC to the mission efficiency. Bayesian optimization is performed to obtain the hyperparameters of the MPC that maximize the mission efficiency. Numerical simulations are performed for both cases using algorithm based on distributed MPC formulation. A performance comparison is provided with the centralized MPC formulation.

Keywords: Task Planning, Manipulation Planning, Motion and Path Planning
Abstract: Rearranging objects on a planar surface arises in a variety of robotic applications, such as product packaging. Using two arms can improve efficiency but introduces new computational challenges. This paper studies the problem structure of revisions{object rearrangement using two arms in} synchronous, monotone tabletop setups and develops an optimal mixed integer model. It then describes an efficient and scalable algorithm, which first minimizes the cost of object transfers and then of moves between objects. This is motivated by the fact that, asymptotically, object transfers dominate the cost of solutions. Moreover, a lazy strategy minimizes the number of motion planning calls and results in significant speedups. Theoretical arguments support the benefits of using two arms and indicate that synchronous execution, in which the two arms perform together either transfers or moves, introduces only a small overhead. Experiments support these claims and show that the scalable method can quickly compute solutions close to the optimal for the considered setup.

Keywords: Deep Learning in Robotics and Automation, Machine learning, Cyber-physical Production Systems and Industry 4.0
Abstract: Digitalization leads to data transparency for production systems that we can benefit from with data-driven analysis methods like neural networks. For example, automated anomaly detection enables saving resources and optimizing the production. We study using rarely occurring information about labeled anomalies into Variational Autoencoder neural network structures to overcome information deficits of supervised and unsupervised approaches. This method outperforms all other models in terms of accuracy, precision, and recall. We evaluate the following methods: Principal Component Analysis, Isolation Forest, Classifying Neural Networks, and Variational Autoencoders on seven time series datasets to find the best performing detection methods. We extend this idea to include more infrequently occurring meta information about production processes. This use of sparse labels, both of anomalies or production data, allows to harness any additional information available for increasing anomaly detection performance.

Keywords: Discrete Event Dynamic Automation Systems, Formal Methods in Robotics and Automation
Abstract: This paper discusses a method to distribute a synthesized supervisory controller for implementation on multiple physical controllers. Dependency structure matrices are used to determine a distribution of a discrete event system. The supervisor is then distributed accordingly, using an existing localization method. Communication delays between the distributed components of a supervisor may affect its behavior, due to changes in the order of events. Therefore, a delay-robustness check is performed and where needed mutex locks are employed to make the distributed supervisor delay-robust. A case study demonstrates the method, and hardware-in-the-loop testing is used to validate the distributed supervisor.

Keywords: Planning, Scheduling and Coordination, Renewable Energy Sources, Modelling, Simulation and Validation of Cyber-physical Energy Systems
Abstract: The issue caused by increasing energy consumption of internet data centers (IDCs) has received critical attention. Hence, it is important to optimize the energy system of IDCs for energy saving and carbon emissions reduction. In order to explore the utilization of hydrogen in IDCs and the advantages of water cooling system, IDCs decarbonized by hydrogen-water-based energy systems are developed in this paper, which makes full use of both electricity and heating energy generated by fuel cells, and renewable energy. The integrated planning-and-scheduling problem is formulated as a two-stage stochastic mixed-integer programming problem to determine the optimal capacity of energy facilities and operation strategies. The developed IDCs are investigated with energy demand and solar radiation in typical days. It is found the developed IDCs are environmental friendly and costeffective. The power usage efficiency can be as low as 1.02 and carbon emission reduction can be up to 59.1% when hydrogen price is U.S. department of energy target price 2/kg.

Keywords: Automation in Construction, Simulation and Animation, AI-Based Methods
Abstract: Smart construction has raised higher automation requirements of construction processes. The traditional construction planning does not match the demands of integrating smart construction with other technologies such as robotics, building information modelling (BIM), and internet of things (IoT). Therefore, more precise and meticulous construction planning is necessary. In this paper, leveraging recent advances in deep Reinforcement Learning (DRL), we design simulated construction environments for deep reinforcement learning and integrate these environments with deep Q-learning methods. We develop reliable controllers for assembly planning for prefabricated construction. For this, we first show that hand-designed rewards work well for these tasks; then we show deep neural policies can achieve good performance for some simple tasks.

Keywords: Physically Assistive Devices, AI and Machine Learning in Healthcare, Rehabilitation
Abstract: Pushrim-activated power-assisted wheels (PAPAWs) are power add-ons that can provide propulsion assistance to manual wheelchair users. These assistive devices operate based on a collaborative control scheme, in which the wheelchair user is an integral part of the control framework. In this work, we aimed to develop a user-centred PAPAW controller that can estimate implicit user intentions (i.e., states of wheelchair motion) and provide adaptive torque assistance accordingly. The proposed framework consists of a data-driven intention-inference module as well as a torque-based controller with adaptive gains. Experiments were conducted to collect kinetic and kinematic data for a variety of common wheelchair activities. Kinetic measurements were used with probabilistic algorithms to develop a user intention estimation model. After the user intent was determined from dynamic user-pushrim interactions, the assistance torque was regulated by adjusting the controller gain to reflect user intent. We evaluated the performance of the intention-based controller in simulation and compared its performance with previously developed PAPAW controllers. We used kinetic and kinematic data from experimental measurements and system identification methods to derive the dynamic model of the user-wheelchair system which was used for evaluation purposes. The outcome of our evaluations showed that using an intention-based controller can reduce unwanted wheelchair deviations from the desired path. User experiments should be conducted to validate these simulation outcomes and verify the clinical relevance of using this intention-based controller.

Keywords: Intelligent and Flexible Manufacturing, Planning, Scheduling and Coordination
Abstract: Multi-axis additive manufacturing enables high flexibility of material deposition along dynamically varied directions. The Cartesian motion platforms of these machines include three parallel axes and two rotational axes. Singularity on rotational axes is a critical issue to be tackled in motion planning for ensuring high quality of manufacturing results. The highly nonlinear mapping in the singular region can convert a smooth toolpath with uniformly sampled waypoints defined in the model coordinate system into a highly discontinuous motion in the machine coordinate system, which leads to over-extrusion / under-extrusion of materials in filament-based additive manufacturing. Moreover, collision may occur when sampling-based collision avoidance is employed. In this paper, we present a motion planning method to support the manufacturing realization of designed toolpaths for multi-axis additive manufacturing. Problems of singularity and collision are considered in an integrated manner to improve the motion therefore the quality of fabrication. Experiments are conducted to demonstrate the performance of our method.

Keywords: Semiconductor Manufacturing, Planning, Scheduling and Coordination, Machine learning
Abstract: The objective of this research is to improve scheduling decisions in high-mix low-volume (HMLV) production environments. Unique characteristics of HMLV semiconductor assembly and testing operations include: (1) Diversified Product Lines: To respond to global competition and different customer needs, manufacturers are providing diversified products to different consumers; (2) Unrelated Parallel Machines: Different machines are oftentimes procured at different capacity expansion stages. While different machines may have similar functions, the latest model oftentimes provides higher production efficiency and better quality. This leads to a production environment with parallel machines of different production characteristics; (3) Incomplete Product-machine Specific Production Data: When there are a wide variety of products produced by unrelated parallel machines, there will be a large number of possible product-machine combinations. Some of the combinations will not have enough data for accurate estimation of production characteristics, such as the required processing time information for scheduling decisions. In order to facilitate efficient and effective scheduling decisions in such HMLV production, a hierarchical prediction method is developed for mixed dataset analysis. The hierarchical prediction method generates missing parameters, such as the required processing time, for subsequent optimization of scheduling decisions. To cope with the inevitable errors in all forecast models, robust scheduling decisions are generated through a stochastic optimization framework. The framework is validated by an industry dataset collected from a leading semiconductor assembly and testing company.

Keywords: Semiconductor Manufacturing, Intelligent and Flexible Manufacturing, Inventory Management
Abstract: This paper investigates two global scheduling strategies for cycle time minimization in semiconductor manufacturing. These global scheduling strategies represented as a linear programming models are compared to a First-in-First out dispatching rule. The first global scheduling strategy is a Push strategy, in which products are pushed to their final operations using high Work-In-Process holding costs on the first operations. The second global scheduling strategy is a Time at Operation strategy, where Work-In-Process quantities that have arrived at different times in an operation are penalized differently. The computational results performed on industrial data using the Anylogic simulation software coupled with IBM ILOG CPLEX show that the Time at Operation strategy minimizes the cycle time while maintaining a high throughput compared to the Push strategy and the simple First-In-First-Out dispatching rule. The paper also shows, when production targets are determined using the Push strategy, products with a large number of operations are prioritized.

Keywords: Semiconductor Manufacturing, Process Control, Optimization and Optimal Control
Abstract: In semiconductor manufacturing the quality and reliability of the products are critical for the companies. The metrology tools used to performs the quality controls are very expensive, therefore the capacity is critical. In this paper we use an indicator that quantify the impact of a scheduling of sampled lots on risk to integrate sampling and scheduling decisions. The resulting problem is NP-hard, we propose a linear programming model and a lagrangian relaxation heuristic to solve industrial instances.

Keywords: Planning, Scheduling and Coordination, Optimization and Optimal Control, Inventory Management
Abstract: In a highly competitive technology industry firms must constantly improve their product portfolio to keep (or gain) market share. This improvement development is done by continually introducing new generations of products into the market in a process that is known as product transition. Furthermore, a firm can also participate in different market niches throughout the industry, dividing the product portfolio into different Product Divisions (PD) offering several different goods at the same time. The product transition process, critical to maintaining competitive advantages in the market, becomes very complex and expensive. Not only due to the sales and marketing efforts of each PD for a proper launch but also in order to be fruitful it requires complete coordination between different agents within the same firm. As the first stage in this research agenda, we present a centralized optimization model to serve as a basis for developing a family of decentralized stochastic decision-making procedures through alternative decompositions and extensions of the subproblems describing the decision problems faced by each agent.

Keywords: Semiconductor Manufacturing, Intelligent Transportation Systems
Abstract: This paper addresses the problem of moving vehicles in an Automated Material Handling System (AMHS)of a semiconductor wafer manufacturing facility network while considering production constraints. An Integer Linear Programming (ILP) model is proposed to enable vehicles to respond as quickly as possible to transport requests. The aim is to minimize waiting times for transport requests. The model has been implemented on Cplex and tested on some instances inspired by industrial environment. Some heuristics have also been proposed, compared with integer linear model and tested in real cases.

Keywords: Semiconductor Manufacturing, Intelligent and Flexible Manufacturing, Hybrid Strategy of Intelligent Manufacturing
Abstract: In semiconductor manufacturing, due to product customization, a cluster tool is frequently required to switch from processing one type of lots to another, leading to ubiquitous lot switching periods, during which consecutive lots are processed concurrently. As a different lot mix often results in different lot completion times, it is challenging to obtain an optimal processing sequence of lots to minimize the makespan for a flexible flow shop with two stages where each stage consists of identical cluster tools. In this paper, an efficient backward earliest-starting strategy is proposed based on which procedures are designed for calculating the completion time of each lot ta¬king into account the different lot mixes. A biased random key genetic algorithm (BRKGA) is proposed to compute high-¬quality schedules using these completion times.

Keywords: Machine learning, AI-Based Methods, Semiconductor Manufacturing
Abstract: In this paper, we adapt the Wavenet architecture for denoising FIB (Focused Ion Beam) image sequences. This method was originally defined to train convolutional neural networks (CNN) with synthetic data for denoising time series. CNN have often shown useful in case of low signal-to-noise ratio. To perform the denoising of image sequences with 1D denoisers, we test several strategies. The first strategy is to extract the signals associated with the evolution of one pixel and to denoise each signal individually. The second (resp. the third) strategy considers each line (resp. column) of each frame as an individual signal. We compare the results obtained with the different strategies and discuss the influence of the architecture for all strategies.

Keywords: Computer Vision in Automation, AI-Based Methods, Failure Detection and Recovery
Abstract: This work applies two state-of-the-art approaches for semantic and instance segmentation of solder voids in X-ray images. Void segmentation is both: an important task in quality and failure analysis of microelectronic components and a challenge to modern computer vision methods, e.g. convolutional neural networks (CNN). We use a CNN named U-Net to distinguish void pixels from background by semantic segmentation. For instance segmentation, we evaluate another CNN, namely Mask-RCNN, which allows the identification of distinct voids instead of a simple binary mask. This approach allows to identify, separate, and evaluate overlapping voids or even voids that lie on top of each other. For the examined dataset, the U-Net outperforms the Mask-RCNN but the work suggests that in a dataset with more overlapping voids, the Mask-RCNN is more favourable.

Keywords: Computer Vision for Manufacturing, Learning and Adaptive Systems, AI-Based Methods
Abstract: Microscopy is commonly used in machining to study the effects of tool wear. In modern tool condition monitoring systems, the analytical capabilities are further enhanced by machine learning, allowing for automated segmentation of the various visible defects. The prevailing challenge, however, is the divergence among different use cases, as the visual properties of cutting tool images are influenced by many domain-specific factors such as the type of the cutting tool, the respective machining process, and the image acquisition unit. Thus, we propose the usage of automated domain adaptation so that existing training data from source domains can be used effectively to train segmentation models for novel target domains, while minimizing the need for newly labelled data. This is achieved through image-to-image translation using generative adversarial networks, which generate synthetic images with similar visual characteristics as the target domain based on existing masks of the source domains. Our validation shows that with as few as ten labelled images from the target domain, a sufficient prediction performance of 0.72~mIoU can be achieved when tested on unseen images from the target domain. This corresponds to a reduction of manual labelling efforts by two-thirds compared to conventional labelling and training methods. Thus, by adapting existing data, prediction performance is increased while expensive data generation is minimized.

Keywords: AI-Based Methods, Deep Learning in Robotics and Automation, Machine learning
Abstract: The robustness of image processing and machine learning algorithms for object and anomaly detection, image segmentation or failure analysis tasks is strongly influenced by the image quality. Image blur is still a problem in microscopy because high quality images of non-planar samples with high resolution are not feasible even with high technical and manual effort. In this paper we evaluate deep neural network models for image deblurring on light optical microscopy data. We present a new image deblurring dataset with sharp ground truth images and a variation of different out-of-focus blur and vibration blur images. We show that image quality enhancement using deep learning methods has great potential in microscopy-based failure analysis. The best method achieved an improvement for the PSNR metric from 31.24 to 35.19, for the SSIM metric from 0.7981 to 0.9472 and for the IoU score from 0.845 to 0.944 on the given test dataset.

Keywords: Autonomous Vehicle Navigation, Motion and Path Planning, Collision Avoidance
Abstract: Autonomous collision-free navigation in dynamic environments is essential in many mobile robot applications. Reinforcement learning has the potential to automate the control design process and it has been successfully applied to path planning and mobile robot navigation. Obtaining effective navigation strategies with reinforcement learning is, however, still very time-consuming, especially in complex continuous environments where the robot can easily get trapped. In this paper, we propose a novel state space definition that includes information about the robot's most recent action. The proposed state variables for the target and nearby obstacles are binary and denote presence (or absence) within a corresponding region in the robot's frame of reference. In addition, we propose two heuristic algorithms that provide the robot with basic prior knowledge about a promising action in each state, which reduces the initial time-consuming blind exploration and thereby significantly shortens training time. We train a robot using our improved Q-learning approach to navigate in continuous environments in a high-fidelity simulator. In a series of experiments, we demonstrate the effectiveness of the proposed approach in terms of training time and solution quality compared to state-of-the-art approaches.

Keywords: Autonomous Vehicle Navigation
Abstract: Autonomous navigation in unknown environment is a big challenge. It needs both good map and effective path planning algorithm for the unknown environment. In this paper, we use sliding mode method to improve the SLAM (simultaneous localization and mapping) with bounded uncertainties. Then we propose a novel path planning method based on the novel SLAM, which uses the genetic algorithm. This novel method takes the advantages of the robustness of the sliding mode method and the good convergence ability of the genetic algorithm. Comparisons with others popular methods are made to show the advantages of the proposed method.

Keywords: Autonomous Vehicle Navigation, Optimization and Optimal Control, Human Factors and Human-in-the-Loop
Abstract: Human-robot interaction, particularly in wheeled mobile robots that can autonomously assist humans to traverse dynamically changing environments is a field of active research. Integrated motion planning and obstacle-avoidance pose a considerable challenge for an autonomous person-following robot (PFR). And, scenarios with intersections and occlusions along the path only increase the complexity in sustained tracking. In this paper, we use model predictive control (MPC) with early-relocation (ER) strategy to formulate a prioritized tracking scheme and implement it for a differential-drive system. Our approach ensures that the target person stays within the field of view (FOV) of the PFR consistently, even while it maneuvers intersections or crowded spots, by adding new locations to its updated path. As trajectory generation in such cases must be incremental to accommodate new information, the use of efficient representations is key. To that end, we build this social representation of following a person directly into the controller itself. MPC can naturally handle such state and input limitations as constraints to solve an on-line optimization at each time step.

A non-linear MPC with ER is thus devised and tested with increasing levels of complexity arising from occlusions due to the map and its dynamic actors. By using 2D simulations, we show that for slow and medium walking speeds of the target person, the controller can plan maneuvers with an adequate margin of over 20 Hz apt for achieving a near real-time person-following behaviour.

Keywords: Intelligent Transportation Systems, Power and Energy Systems automation, Autonomous Vehicle Navigation
Abstract: Autonomous unmanned aircraft (UA) are currently gaining importance in various areas of application. In addition to their use in the field of surveying, inspection or fluid application, the focus is also on the transport of goods and the monitoring of factory premises.

However, decisive disadvantage of commonly used UA is their short flight time, which typically ranges between approx. 10 and 45 minutes, depending on the design and payload. For the automated use of UA, it is not practical to manually change the battery between individual flights but automated solutions are required. The automated charging methods presented so far have in common that the UA approaches the landing platform vertically or at an angle from above and stands on the landing platform after completion.

In this paper, we present a novel solution for the overhead charging of UA. Thereby the UA docks vertically to the charging station from below and is locked in the station. This saves floor space and allows the use of sensors attached to UA for information acquisition and environmental monitoring while charging. In addition to the charging station and the associated docking device attached to the UA, a camera-based method for localizing the charging station and for precise docking is presented. Finally, the docking and undocking process is evaluated in a series of experiments.

Keywords: Autonomous Vehicle Navigation, Task Planning, Industrial and Service Robotics
Abstract: The Vehicle Routing Problem (VRP) is the combinatorial optimization problem of designing routes for vehicles to visit customers in such a fashion that a cost function, typically the number of vehicles, or the total travelled distance is minimized. The problem finds applications in industrial scenarios, for example where Automated Guided Vehicles run through the plant to deliver components from the warehouse. This specific problem, henceforth called the Electric Conflict-Free Vehicle Routing Problem (CF-EVRP), involves constraints such as limited operating range of the vehicles, time windows on the delivery to the customers, and limited capacity on the number of vehicles the road segments can accommodate at the same time. Such a complex system results in a large model that cannot easily be solved to optimality in reasonable time. We therefore developed a compositional algorithm that breaks down the problem into smaller and simpler sub-problems and provides sub-optimal, feasible solutions to the original problem. The algorithm exploits the strengths of SMT solvers, which proved in our previous work to be an efficient approach to deal with scheduling problems. Compared to a monolithic model for the CF-EVRP, written in the SMT standard language and solved using a state-of-the-art SMT solver the compositional algorithm was found to be significantly faster.

Keywords: Autonomous Vehicle Navigation, Reactive and Sensor-Based Planning, Probability and Statistical Methods
Abstract: This paper presents a method for 3D robot localization in an airport environment using an enhanced adaptive Monte Carlo localization algorithm with sensor input from a multi-beam lidar. The occupancy grid is computed based on the known geometry of various airplane types. By inserting adaptive particle noise, an estimated global pose can be obtained reliably in stationary conditions with a low number of initial particles. As the probabilistic particle filter converges, the particle noise and the number of particles are reduced. Robot odometry is used to propagate the candidate particles when moving. The algorithm has been implemented within the Robot Operating System (ROS) framework and can run in real-time on a low-power computing device on the robot. Comparison of the numerous enhancements are shown in simulation. The results have been validated in practice on multiple airplanes at two airports showing good performance.

Keywords: Modelling, Simulation and Optimization in Healthcare, Diagnosis and Prognostics, AI-Based Methods
Abstract: Anomaly detection algorithms have vast applications, from fraud detection in business transactions to rare pattern detection in a production line to help prevent machinery failures. The availability of quantitative clinical data makes a compelling case for using anomaly detection algorithms in clinical settings, for instance, to help prevent diagnosis errors. This work evaluates the feasibility of using Isolation Forest algorithm for detection of spikes in surface electromyography (sEMG) of biceps and muscle resistive force in upper limb spasticity datasets. Results show that the anomaly detection in sEMG data is a good predictor for the occurrence of catch. It could be deployed in rehabilitation robotic systems for injury prevention by linking the anomaly detection to the actuation module exerting force in the system.

Keywords: Modelling, Simulation and Optimization in Healthcare, AI and Machine Learning in Healthcare, Clinical and Operational Decision Support
Abstract: This paper applies topological data analysis (TDA) techniques to investigate the nature of complex high-dimensional data by extracting global shape information (patterns) and gaining novel insights from them. The objective is to characterize miscoding behaviors, identify reasons for miscoding behaviors, and select specific groups of subjects for which the health records are worth giving an additional review. Our method combines a TDA technique and an optimization-based model to provide a geometric representation of inter-related hospital stays while permitting the censoring of miscoded subjects by preferentially selecting subgroups with more coding errors. Through the proposed method, we successfully identified and validated four distinct subtypes of miscoding behaviors that traditional methodologies fail to find. Furthermore, with only 20% of the subjects reviewed, the proposed approach reduces coding errors by 64% of the whole population. Experimental results indicate that the proposed method is promising and can reduce coding errors efficiently, thereby eliminating the negative impacts caused by hospital miscoding.

Keywords: Modelling, Simulation and Optimization in Healthcare, Medical Robots and Systems, Additive Manufacturing
Abstract: The additive manufacturing technology greatly shortens the design and fabrication period of a complex robotic system, as the entire system can be once printed in a short time without complicated assembly process. To keep up with the fast development of this new technology, an efficient tool for constructing compact and 3D-printable surface models is highly desirable. In this paper, we present a novel boundary-representation-based geometry modeling language, the Solid Geometry Coding Language (SGCL), for the efficient surface modeling of 3D-printable medical robots. The proposed language is implemented in MATLAB, based on our previously developed modeling toolbox, the Solid Geometry Library. The basic data structures and modeling syntax of the language are illustrated in detail. To demonstrate the performance of the SGCL language, a modeling case as well as the 3D-printed prototype is presented. Furthermore, the proposed language could also be used as an open platform for developing optimization-based automatic synthesis methods for 3D-printed medical instruments and robots.

Keywords: Modelling, Simulation and Optimization in Healthcare, Health Care Management, Clinical and Operational Decision Support
Abstract: Hospital readmissions are common and very costly to the healthcare sector. Besides, readmission is a burden to the patient as well as to the healthcare system. Moreover, a high rate of hospital readmissions is increasingly viewed as a marker of poor quality care . Hospital readmission prediction is characterized by highly imbalanced data and a large number of features that complicate mathematical modeling. An efficient approach that integrates various innovative components is proposed including : 1) a Tabu search approach for optimal feature selection, 2) an optimization-based classification model based on discriminant analysis (DAMIP) 3) undersampling to achieve a more balanced training data set, and 4) random sampling to cope with the large patient population and to speed up the feature selection.The proposed classification model offers the possibility to leave certain entities unclassified, in case of a highly uncertain classification. The model has been tested on French healthcare data including patient characteristics, diagnosis information, and medical procedure information on 15,000 hospital admissions. Varying readmission intervals were tested (30 days/90 days/180 days) and prediction results were compared to those of classic machine learning algorithms. The experiments show a performance improvement of classification by applying Tabu search for feature selection, where comparison is done based on F1-score. This method achieves an F1-score of over 0.605 whereas the F1-scores of competing techniques are under 0.410. Regarding practical applications, such prediction approach will be used by practitioners to focus more on care organization for patients at risk of readmission.

Keywords: AI and Machine Learning in Healthcare, Health Care Management, Planning, Scheduling and Coordination
Abstract: Hospital beds management is critical for the quality of patient care, while length of inpatient stay is often estimated empirically by physicians or chief nurses of medical wards. Providing an efficient method for forecasting the length of stay (LOS) is expected to improve resources and discharges planning. Predictions should be accurate and work for as many patients as possible, despite their heterogeneous profiles.

In this work, a LOS prediction method based on deep learning and embeddings is developed by using generic hospital administrative data from a French national hospital discharge database, as well as emergency care.

Data concerned 497626 stays of 304931 patients from 6 hospitals in Lyon, France, from 2011 to 2019. Results of a 5-fold cross-validation showed an accuracy of 0.73 and a kappa score of 0.67 for the embeddings method. This outperformed the baseline which used the raw input features directly.

Keywords: Automation in Life Science: Biotechnology, Pharmaceutical and Health Care, Diagnosis and Prognostics, Smart Home and City
Abstract: World population ageing causes an important increase of people needing specific health care and monitoring. Dedicated institutions exist, but most of the elderly prefer to keep their autonomy for economic and personal reasons. To ensure a good quality of life and health to this population, Health at Home (HaH) solutions are explored. Many works focus on monitoring smart home inhabitant behavior to detect changes which might be due to health problems. These approaches are efficient to detect accident or short-term diseases such as a cold or influenza but tend to detect too tardily diseases which provoke slow declines in behavior. This is a problem as the elderly are likely to suffer from such troubles and early detection allows for better diagnosis and may help to prevent or reduce future worsening. In this paper, a novel approach for the detection of long-term behavior changes is introduced. It focuses on activity duration as this indicator is influenced by most diseases and gives clear information about the inhabitant's health status. This paper proposes data forecasting to detect future anomalies to assess the existence of evolution in the current behavior. Information is sent to medical staff to refine their prognostic and adapt their treatment or call for a medical appointment. A case study based on a real smart home simulating a worst-case scenario attests for the efficiency of the approach and its resilience.

Keywords: AI-Based Methods, Computer Vision in Automation, Deep Learning in Robotics and Automation
Abstract: Mobile robots have the potential to be used in many outdoor tasks, such as search and rescue, patrolling, and delivery services. To enable robots to safely navigate through outdoor environments, it is important to analyse the terrain. We present a novel approach using Semantic Terrain Segmentation (STS), which relies on an adaptation of Deeplabv3+. Our goal is to segment the terrain according to different height thresholds, using only aerial RGB images. We collected and labelled a dataset, Vale, consisting of four categories based on the traversability constraints of three types of mobile robots: wheeled, tracked, legged and non-traversable. We trained our deep convolutional neural network (DCNN) on Vale, with transfer learning using a network pre-trained on Cityscapes. Our findings suggest that the proposed DCNN approach can identify and differentiate different height thresholds in the terrain, i.e. can segment based on mobile robot traversability.

Keywords: Computer Vision in Automation, Motion and Path Planning, Agent-Based Systems
Abstract: Recovering the trajectory of a moving camera from image sequences usually generates a set of discrete camera poses. Applications such as view synthesis, sensor data fusion and interactive simulations, however, often require a continuous trajectory. In this work an algorithm for estimating the control points of a cubic Bézier curve is presented and evaluated on the impact of noisy measurement data. With the estimation results a continuous-time representation of the camera’s trajectory is recovered and a pose interpolation between the noisy discrete camera poses is realizable. Different levels of noise on translation, rotation and velocity are analysed, whereby the use of straight line trajectories leads to special configurations. The contributions of this paper are as follows: (1) the absolute trajectory error compared on various noisy parameters in characteristic maps (2) the continuous-time trajectory generation combined with special configurations using straight line trajectories and (3) the source code of implementation and measurement examples are released, in order to make a reconstruction of the results and application in related research possible.

Keywords: Computer Vision in Automation, AI-Based Methods, Factory Automation
Abstract: To make sure the quality and reliability of lithiumion batteries(LIBs) and improve detection speed, developing automatic defects detection to take the place of manual detection has been a general trend in the quality control lines of LIBs. In this paper, a detection method based on X-ray technology and convolutional neural network(CNN) is proposed for internal wrinkles detection in LIBs. Besides, for reducing false positive rate, loss fuction is modified by adding penalty coefficient when training CNN model. The proposed method has a nice performance in accuracy and false positive rate, and satisfies industrial requirements, and has been applied in the quality control of Lithium-ion battery production lines

Keywords: Computer Vision in Automation, Sensor Fusion, Sensor-based Control
Abstract: A novel concept for visual-lengthwise configuration self-estimator (VLC-SE) of continuum robots is presented, using multiple monocular cameras mounted to each end of segments and lengths encoder as input sensors. The proposed approach grounds the improved lengths based kinematic model with Piecewise Polynomial Curvature (PPC) hypothesis, which ensures accurate modelling and avoids the flaws – as discontinuities and singularities. Meanwhile, we discussed the observability of the improved model. We propose to enhance perception by the vision of each segment end, which comes to the concept of Flexible Multi-Camera Bundle Adjustment (FMC-BA). We validate the performance of our approach on the data collected on a snake-like continuum robot. We also share the first continuum robot datasets: CoRo, including vision and arc lengths data, to promote further research. (https://cutt.ly/CoRo_dataset).

Keywords: Computer Vision in Automation, Deep Learning in Robotics and Automation, Computer Vision for Manufacturing
Abstract: Printed Circuit Board Assembly (PCBA) plays an important role in 3C (Computer, Communication and Consumer Electronics) industry. The object detection of electronic components is the key technology for flaw inspection and quality control of PCBA. YOLOv3 has been widely used in object detection. Image pre-processing can eliminate some distractions and prominent some feature to improve the detection accuracy. The tailored YOLOv3 has a smaller computational cost and transfer learning can accelerate the training speed and improve inspection accuracy.

Keywords: Computer Vision in Automation, Deep Learning in Robotics and Automation, Probability and Statistical Methods
Abstract: In this paper, we introduce a new technique that combines two popular methods to estimate uncertainty in object detection. Quantifying uncertainty is critical in real-world robotic applications. Traditional detection models can be ambiguous even when they provide a high-probability output. Robot actions based on high-confidence, yet unreliable predictions, may result in serious repercussions. Our framework employs deep ensembles and Monte Carlo dropout for approximating predictive uncertainty, and it improves upon the uncertainty estimation quality of the baseline method. The proposed approach is evaluated on publicly available synthetic image datasets captured from sequences of video.

Keywords: Mechanism Design in Meso, Micro and Nano Scale, Force and Tactile Sensing, Automation at Micro-Nano Scales
Abstract: Stimulus-responsive polymer sensors are important components of micro/nano-mechanical systems (M/NEMS), which are widely used in many frontier fields. As an important part of smart materials, the volume, mass, or elasticity of pH-sensitive polymers can shift with pH values, which can be used in many fields such as biology, chemistry and micro/nano electromechanical system. However, few studies on the micromechanical properties of smart materials have been reported until now. In this paper, a comparative study of the single-molecule mechanical elasticity of pH-sensitive polymeric polyacrylic acid (PAA) at pH change was performed using atomic force microscopy-based single-molecule force spectroscopy (SMFS). The results show that the single-chain conformation of PAA undergoes from collapse to full extension with increasing pH and the energy difference between different conformations is obtained, which leads to a novel design concept of a molecular motor (switch). It is expected that our study can provide a theoretical basis and data support for the design of new polymers and smart sensors with multiple responses.

Keywords: Computer Vision in Automation, Computer Vision for Manufacturing
Abstract: The estimation of 6D poses of industrial parts is a fundamental problem in smart manufacturing. Traditional approaches mainly focus on matching corresponding key point pairs between observed 2D image and 3D object model via hand-crafted feature descriptors. However, key points are hard to discover from images when the parts are piled up in disorder or occluded by other distractors, e.g., human hands. Although the emerging deep learning-based methods are capable of inferring the poses of occluded parts, the accuracy is not satisfactory largely due to the loss of spatial resolution from multiple downsampling operations inside convolutional neural networks. To overcome this challenge, this paper proposes a 6D pose estimation model consisting of a pose estimator and a pose refiner, by leveraging High-Resolution Network as the backbone network. Experiments are further conducted on a dataset of industrial parts to demonstrate its effectiveness.

Keywords: Diagnosis and Prognostics, Machine learning, Semiconductor Manufacturing
Abstract: Fault detection and classification (FDC) is important for semiconductor manufacturing to monitor equipment’s condition and examine the potential cause of the fault. Each equipment in the semiconductor manufacturing process are often accompanied by a large amount of sensor readings, also called status variable identification (SVID). Identifying the key SVIDs accurately can make it easier for engineers to monitor the process, and maintain the stability of the process and wafer productive yields. This paper proposes using the random forests algorithm to analyze the importance of SVIDs of equipment sensors, automatically filters the key SVID by using k-means, and integrates various machine learning methods to verify the key SVIDs and identify key processing time and steps. Upon the key parameters are identified, the key processing time and steps are investigated subsequently. The ensemble models constructed on K-nearest neighbors (kNN) and naïve Bayes classifiers are presented for classifying wafers as normal or abnormal. Data visualization of multi-dimensional key SVIDs is performed by using t-distributed stochastic neighbor embedding (t-SNE) to create a graphical aid in FDC for the process engineer. An empirical study is conducted to validate the proposed data-driven framework for fault detection and diagnostic. The experimental results demonstrate that the proposed framework can detect abnormality effectively with highly imbalanced classes and also gain insightful information about the key SVIDs and corresponding key processing time and steps.

Keywords: Additive Manufacturing, Probability and Statistical Methods
Abstract: The need for estimating three-dimensional (3D) information based on two-dimensional (2D) images has been increasing in numerous fields. It is essential in quality assessment, quality control, and process optimization. However, all the existing methods have not considered the data truncation issue, which is commonly faced in metrology. This paper proposes a new statistical approach to infer size distribution, volume number density of 3D particles based on 2D cross-sectional images with data truncation considered. In order to estimate the size distribution, a linkage is established between 3D particles and 2D observations with the existence of data truncation. Subsequently, this paper derives the likelihood function of 2D observations and an efficient Monte Carlo EM algorithm is developed to estimate the parameters of size distribution. In addition, an explicit relationship between the 3D and 2D particle number densities is established and leveraged to estimate the volume number density and volume fraction. The effectiveness of the proposed method is demonstrated through both simulation study and real case studies in metal additive manufacturing (AM) and metal-matrix nanocomposites (MMNCs) manufacturing.

Keywords: Planning, Scheduling and Coordination, Intelligent and Flexible Manufacturing, Semiconductor Manufacturing
Abstract: Manufacturing system is a complicated and dynamic system which is affected by a variety of uncertain production disturbances. This paper addressed the dynamic scheduling problem of smart production systems and proposed an autonomous and adaptive scheduling method based on Multi- Agent Deep Deterministic Policy Gradients (MADDPG), a multi-agent deep reinforcement learning algorithm, by making full use of real-time data resources and edge computing ability. Different from the traditional dynamic scheduling methods, the proposed method has the characteristics of centralized training, distributed autonomous decision-making and online updating.

Keywords: Manufacturing, Maintenance and Supply Chains, Machine learning, Zero-Defect Manufacturing
Abstract: In the production of cold-rolled strip products, strip breakage is one of the most common failures during the cold rolling process. However, the existing prediction models on strip breakage use the conventional sliding window algorithm to process the time series data collected from the actual production, resulting in a massive amount of non-informative data, which increases the computational cost for data-driven modelling. In order to tackle this issue, this article proposed a sliding window filter method to optimise the data pre-processing of the strip breakage. Firstly, based on the existing research and understanding of strip breakage, the data characteristics in the process of strip breakage was analysed. Based on the analysis, sample variance (VAR) and length normalised complexity estimate (LNCE) were chosen to determine how informative the time window was related to strip breakage. Secondly, compared with the conventional sliding window approach, the sliding windows were classified through a filter using VAR and LNCE. Thirdly, the filtered data was fed into the Recurrent Neural Network (RNN) for strip breakage modelling. An experimental study based on actual production data collected by a cold-rolled strip manufacturer was conducted to verify this method's effectiveness. The results show that pre-processing data using the sliding window filter decreases the model's computational cost.

Keywords: Logistics, Planning, Scheduling and Coordination
Abstract: This paper studies an integration of the forward products flow, as in the vehicle routing problem with cross-docking (VRPCD), and the reverse products flow, as in the VRP with reverse cross-docking (VRP-RCD), namely the VRP with forward and reverse cross-docking (VRP-FRCD). It is modelled as a mixed-integer linear programming model that covers a four-level supply chain network, including suppliers, retailers, outlets, and a cross-dock facility. The objective is to determine the appropriate number of vehicles used to complete the delivery process, together with the route sequence of every vehicle, such that the total fixed vehicle and distance-related costs are minimized. In order to solve the problem, a two-phase matheuristic is developed. The first phase focuses on finding as many route combinations as possible. This is implemented by an adaptive large neighborhood search (ALNS) algorithm. Subsequently, a set partitioning formulation is formulated and solved in the second phase to determine the best combination over all routes found in the first phase that minimizes the total cost. The performance of the matheuristic in solving the newly developed benchmark instances is compared against those of CPLEX and a pure ALNS algorithm. Experimental results suggest that the matheuristic is more powerful and beneficial than both ALNS and CPLEX to obtain high-quality solutions within an acceptable computational time.

Keywords: Planning, Scheduling and Coordination, Intelligent Transportation Systems
Abstract: This paper addresses decision making for networked autonomous vehicles in mobility on demand (MoD) systems. An optimization formulation, termed Pick-up, Delivery, and Rebalancing Problem with Time Window (PDRPTW), that simultaneously account for the routing & scheduling of the vehicles in response to existing service requests and the rebalancing of them for future requests is presented in the state-space-time network representation of the problem. Then, the alternating direction method of multipliers (ADMM) method is adopted to effectively solve this optimization problem in a distributed manner. The ADMM framework allows for decomposition of the problem into minimization of operational cost of individual vehicles and minimization of imbalance cost in the stations; the method leads to consensus upon the routing and waiting plans of the vehicles. Numerical examples demonstrate the efficacy and the benefits of the proposed distributed algorithm compared to other multi-vehicle task allocation schemes.

Keywords: Space Robotics and Automation, Planning, Scheduling and Coordination
Abstract: Cable-driven hyper-redundant manipulators (CDHMs) have been widely applied in aerospace, surgery, or other fields to complete tasks in narrow and multiobstacles environments. Maneuverability is an important metric for CDHMs and is more difficult to analyze than traditional industrial robots because of their more complicated kinematics. So far, the proposed methods for evaluating the manipulability of cable-driven robots are similar to those of traditional industrial robots in that they focus on the joints and ignore the cables. However, these methods are flawed for cable-driven robots because they can’t guarantee the dexterity of the true driving space (i.e., cable space). In this paper, a tailored analysis tool is proposed, which contains two core metrics, namely, cable velocity manipulability ellipsoid (CVME) and cable-driven cost index (CDCI), to evaluate CDHMs’ manipulability of cables, joints, and the end-effector. Firstly, the Jacobian matrix mapping from the end-effector velocity to the cable velocity is solved based on minimizing the norm of the cable velocity. Then the CVME and CDCI are defined through this Jacobian matrix. Finally, we show how to perform a specific analysis of the manipulability of CDHMs using CVME and CDCI. The simulation results demonstrate that the CVME and CDCI are reasonable and effective.

Keywords: Planning, Scheduling and Coordination, Petri Nets for Automation Control, Reinforcement
Abstract: We propose a reinforcement learning algorithm for an integrated production and material handling scheduling problem. We model the problem with a timed Petri net (TPN) and define states, actions, and rewards of reinforcement learning using the TPN components. The experimental results show that the proposed algorithm performs better than other dispatching rules.

Keywords: Planning, Scheduling and Coordination, AI-Based Methods
Abstract: Avoiding collisions is the core problem in multi-agent navigation. In decentralized settings, when agents have limited communication and sensory capabilities, collisions are typically avoided in a reactive fashion, relying on local observations/communications. Prominent collision avoidance techniques, e.g. ORCA, are computationally efficient and scale well to a large number of agents. However, in numerous scenarios, involving navigation through the tight passages or confined spaces, deadlocks are likely to occur due to the egoistic behaviour of the agents and as a result, the latter can not achieve their goals. To this end, we suggest an application of the locally confined multi-agent path finding (MAPF) solvers that coordinate sub-groups of the agents that appear to be in a deadlock (to detect the latter we suggest a simple, yet efficient ad-hoc routine). We present a way to build a grid-based MAPF instance, typically required by modern MAPF solvers. We evaluate two of them in our experiments, i.e. PUSH AND ROTATE and a bounded-suboptimal version of CONFLICT BASED SEARCH (ECBS), and show that their inclusion into the navigation pipeline significantly increases the success rate, from 15% to 99% in certain cases.

Keywords: Planning, Scheduling and Coordination
Abstract: In order to let multiple robot manipulators cooperatively complete a sequence of tasks in a shared workspace under task execution uncertainty, this paper proposes a multi-robot task allocation framework for constantly assigning tasks to robots, while the interference among concurrent robot motions is account for. An online sequential task assignment method is presented, which decouples the time-extended problem into a sequence of synchronous and asynchronous instantaneous assignment sub-problems. This renders the approach capable of reacting to task execution uncertainties in real-time. A one-step-ahead simulation method is employed to reduce the idle time of robots and improve task completion efficiency. Each instantaneous assignment sub-problem is modeled as an optimal assignment problem with variable utility and solved by a branch-and-bound algorithm, with which multi-robot motion coordination is integrated. Experimental results conducted with three Franka-Emika Panda arms show that these can cooperatively complete all tasks without collision and little waiting time. Simulations with larger multi-robot systems show that the approach scales linearly with the number of robots.

Keywords: Process Control, Big-Data and Data Mining, AI-Based Methods
Abstract: Polysilicon was prepared by Chemical Vapor Deposition (CVD) process in the Siemens reactor. We can use model predictive control (MPC) to control the CVD reactor, in which the prediction of target variables is very important. We propose to use neural networks to predict the key variables of the CVD reactor. The performance measure for the model is the mean standard error (MSE). The experimental results show that the MSE of the target parameter prediction is 0.00612, and the algorithm achieves the prediction effect accurately. We believe the proposed approach is of great interest for the operation of the polysilicon CVD process.

Keywords: Big-Data and Data Mining
Abstract: Abstract: Members of book supply chain have begun to pay attention to big data and explore ways and methods to use it. However, for the traditional publishing knowledge supply chain, facing wholesale pricing and agency pricing models about book, they want to know the condition of BDIS Investment and how to price. To solve these problems, a book supply chain with one publisher and one retailer was chosen. Moreover, the publisher releases electronic version and paper version books simultaneously. Considering wholesale pricing and agency pricing models, we built and analysed benefit models in the proposed four situations. Findings: 1) if publisher and retailer want to gain more benefits. They should try their best to attract the value of Big Data information and reduce the value of the BDIS optimization coefficient. 2) If publisher and retailer have invested in BDIS, adopting wholesale pricing model will help members gain more benefits.

Keywords: Big-Data and Data Mining, Intelligent Transportation Systems, Machine learning
Abstract: Passenger flow forecasting is a very critical task for the daily operations of metro system. The rapid development of deep learning methods offers us an opportunity to give an end-to-end solution to system level prediction. However, complex spatial-temporal correlation makes it quite challenging. Existing studies cannot take full advantage of human knowledge and external information. Meanwhile, they also tend to model spatial and temporal dependence separately instead of modelling simultaneously, which may lead to the loss of latent dependence. To bridge the research gap, in this study, we propose a well-designed transformer based spatial-temporal fusion network (TSTFN). To cooperate with different types of external information and give additional insights, we first use multiple pre-defined graph structures to construct multi-view GCN for spatial dependence modelling. Then we propose a novel spatial-temporal synchronous self-attention layer to model spatial and temporal correlation simultaneously. Experiments show TSTFN outperforms other state-of-the-art deep learning based methods on both long-term and short-term tasks. The effectiveness of crucial components has also been verified by using ablation study and analysis.

Keywords: Big-Data and Data Mining, Machine learning
Abstract: A community structure is a fundamental property of complex networks and its detection plays an important role in exploring and understanding such networks. Due to its great interpretability, a symmetric and non-negative matrix factorization (SNMF) model is frequently adopted to perform community detection tasks. However, it adopts a single latent factor (LF) matrix to construct the approximation of a given undirected matrix to ensure its absolute symmetry at the expense of shrinking its solution space. In this paper, we propose a symmetry-constrained NMF (SCNMF) method. We model the approximate symmetry of an undirected network by introducing an equality-constraint on LF matrices into an NMF framework. Besides, we use graph-regularization to extract the features regarding the intrinsic geometric structure of a network. Extensively empirical studies on six real-world social networks from industrial applications demonstrate that the proposed SCNMF-based detector achieves higher accuracy for community detection than state-of-the-art models.

Keywords: Big-Data and Data Mining, Machine learning
Abstract: Large-scale undirected weighted networks are frequently encountered in real applications. They can be described by a Symmetric, High-Dimensional and Sparse (SHiDS) matrix, whose sparse and symmetric data should be addressed with care. However, existing models either fail to handle its sparsity effectively, or fail to correctly describe its symmetry. For addressing these issues, this study proposes an Alternating-direction-method-of-multipliers-based Symmetric Nonnegative Latent Factor Analysis (ASNL) model. Its main idea is three-fold: 1) introducing an equality constraint into a data density-oriented learning objective for a flexible and effective learning process; 2) confining an augmented term to be data density-oriented to enhance generalization the model’s ability; and 3) utilizing the principle of alternating-direction-method of multipliers to divide a complex optimization task into multiple simple subtasks, each of which is solved based on the results of previously solved ones. Empirical studies on two SHiDS matrices demonstrate that ASNL obtains higher prediction accuracy for their missing data than state-of-the-art models with competitive computational efficiency.

Keywords: Big-Data and Data Mining, Machine learning
Abstract: A dynamically weighted directed network (DWDN) is frequently encountered in various big data-related applications like a terminal interaction pattern analysis system (TIPAS) concerned in this study. It consists of large-scale dynamic interactions among numerous entities. Moreover, as the involved entities increase drastically, it becomes impossible to observe their full interactions at each time span, making a corresponding DWDN high-dimensional and incomplete. However, it contains vital knowledge regarding involved entities’ behavior patterns. To extract such knowledge from DWDN, this paper proposes a novel Alternating direction method of multipliers (ADMM)-based Nonnegative Latent-factorization of Tensors (ANLT) model. It adopts two novel ideas: a) building a data density-oriented augmented Lagrangian function for efficiently handling a tensor’s incompleteness and nonnegativity; and b) splitting an optimization task in each iteration into an elaborately designed subtask series where each one is solved based on the previously solved ones following the ADMM principle to achieve fast model convergence. Experimental results on two large-scale DWDNs from a real TIPAS demonstrate that the proposed ANLT model outperforms state-of-the-art models significantly in both computational efficiency and prediction accuracy when addressing missing link prediction on DWDW.

Keywords: Robot Networks, Motion and Path Planning, Swarms
Abstract: This work focuses on the wireless connectivity of multi-agent lunar robotic systems and how it can be preserved during large-scale lunar exploration missions. In particular, we consider in this work the connectivity of systems composed of a single lunar module and several micro-rovers performing coordinated area coverage exploration tasks. To this end, we adopted a deterministic model for lunar radio propagation to predict the status of point-to-point communication links for agents operating on the moon. We then used this information to build a communication graph for the lunar micro-rover network. The Fiedler value, a metric derived from algebraic graph theory, was then utilized for evaluating the system's evolving network connectivity as the micro-rovers explore finite regions on the lunar surface. Simulations involving a network consisting of a single fixed lunar module and three mobile micro-rovers were performed to illustrate how the rovers' basic mobility can cause disruptions in network connectivity. Results of the simulations show that the overall connectivity of lunar multi-rover networks can be maintained by imposing constraints on the rovers' motion.

Keywords: Motion and Path Planning, Task Planning
Abstract: Contact-based motion planning for manipulation, object exploration or balancing often requires finding sequences of fixed and sliding contacts and planning the transition from one contact in the environment to another. However, most existing algorithms concentrate on the control and learning aspect of sliding contacts, but do not embed the problem into a principled framework to provide guarantees on completeness or optimality. To address this problem, we propose a method to extend constraint-based planning using contact transitions for sliding contacts. Such transitions are elementary operations required for whole contact sequences. To model sliding contacts, we define a sliding contact constraint that permits the robot to slide on the surface of a mesh-based object. To exploit transitions between sliding contacts, we develop a contact transition sampler, which uses three constraint modes: contact with a start surface, no contact and contact with a goal surface. We sample these transition modes uniformly which makes them usable with sampling-based planning algorithms. Our method is evaluated by testing it on manipulator arms of two, three and seven internal degrees of freedom with different objects and various sampling-based planning algorithms. This demonstrates that sliding contact constraints could be used as an elementary method for planning long-horizon contact sequences for high-dimensional robotic systems.

Keywords: Motion and Path Planning
Abstract: In this paper, we propose a multi-UAV path planning strategy for fixed wing UAVs to provide convoy protection to a ground vehicle moving over a hilly terrain. As the ground target motion is in 3D, we have developed a Model-Predictive control based framework for tracking the convoy taking the terrain restriction, kinematic constraints of the UAV and the camera field-of-view into account. Due to these constraints, it may not be possible for a single UAV to track the convoy continuously and hence we develop a cooperative multi-UAV framework with two UAVs to ensure continuous tracking. The simulation results show that the proposed framework adequately tracks the target without losing the line-of-sight of the convoy.

Keywords: Motion and Path Planning, Formal Methods in Robotics and Automation, Planning, Scheduling and Coordination
Abstract: Increasingly, various types of deep neural networks are used in physical applications where safety is a primary concern. Unfortunately, existing models are still not robust enough to safely be used in all situations. The work performed in this paper shows definitively, that input corruptions can be mitigated by a properly designed fusion network which utilizes multiple sensing modalities. The fusion network created for this paper consists of 2 input branches, a 2D CNN branch and a 3D point cloud branch. The 2D CNN branch utilizes a ResNeXt feature extractor and the 3D point cloud branch utilizes a dgcnn feature extractor. Experiments show that the fusion network achieves state-of-the-art results.

Keywords: Motion and Path Planning, Energy and Environment-aware Automation
Abstract: This paper presents a fast, energy-efficient, and low computational cost traversal solution on sloped terrain. The use of grid-based search algorithms requires high computational power and takes a long time because almost every point on the map is visited. An approach that does not depend on the global map but can also navigate towards the target can be presented as a new solution. A cost map for motion planning using depth field and color image data is formed in real-time. The proposed motion planning algorithm, named SAFARI, utilizes four cost layers to efficiently evaluate its surroundings. To reduce the computational overhead, only select features are evaluated and the rover's motion planning cycle speed is increased. SAFARI has been tested against path planning alternatives and has also been proven to work with simulations and field tests. This concept is expected to be used in space applications and cave exploration tasks.

Keywords: Foundations of Automation, Motion and Path Planning
Abstract: Smoothness of mobile and vehicle navigation has become relevant to ensure the safety and the comfortability of riding. The robotics community has been able to render smooth trajectories in mobile robots by using non-linear optimization approaches and well-known fairness metrics considering the curvature variations along the path. In this paper, we evaluate the possibility of computing smooth paths from input reference trajectories by using higher order non-linear fairness functionals. Our approach is potential to enable the generation of simple and computationally-efficient path planning and smoothing for navigation in mobile robots.