The design of decentralized, distributed embodied systems benefits from self-organization (e.g., resilience, scalability, and adaptivity to dynamic environments). However, most studies in self-organization on physical platforms (e.g., swarm robotics) were performed within lab environments. The real world comes with severe requirements, calling for more general methodologies, design method standardization, and validation/comparability via benchmarking tool-sets. With this Research Topic, we want to collect, benchmark, and survey novel approaches to push self-organization towards real-world applications, focusing on embodied artificial systems, such as multi-robot, cyber-physical, and socio-technical systems.
We expect research papers on how we can kick our robots out of the lab, pushing towards a novel ‘field swarm robotics’; other relevant fields are self-organizing cyber-physical systems in the wild and distributed systems for radically novel applications, and theoretical or practical studies on novel methods or applications for self-organization in the real world. We also seek for survey and perspective papers on requirements, methodology, and applications for real-world systems. We want to highlight the following requirements, methods, and benchmarking aspects.
Requirements
? Resilience: studies on decentralized systems too often take robustness for granted because of redundancy. Redundancy, however, is only a precondition for robustness. We require novel approaches that make embodied systems robust against random faults and resilient to intentional Byzantine agents. For example, revisit classical coordination algorithms and check how to (1) break them, (2) fix them, and (3) create formal quantitative measures.
? Scalability: Scalability poses stronger requirements in open systems where agents join and leave at runtime. Robots may break or be added to the system, requiring to dynamically scale and reconfigure to the system size.
? Adaptivity: Biological self-organizing systems are known to be adaptive. However, in artificial systems, it is an under-researched requirement. Real-world systems are exposed to unanticipated situations with unavoidable uncertainty, requiring online adaptations. The self-organizing embodied system may be heterogeneous and has to deal with tasks and environments of high complexity.
Methods
? Methods for radically new applications that are driven by real-world requirements
? Learning: We require novel methods of offline/online learning for increased adaptivity in self-organizing systems in open environments. We encourage contributions of online distributed (evolutionary) learning, including embodied evolutionary robotics. We also welcome the application of deep reinforcement learning to distributed embodied systems that resolve the challenge of sparse data. Contributions should address the reality gap and the challenge of designing resource-efficient learning methods.
? Design, modelling, deployment: A general methodology is required for the design of collective systems. Key design challenges are: defining the initial requirements of hardware/software, designing interactions among the cyber-physical systems, and validating the deployed system.
Benchmarking
? Benchmarking standards: Comparability of approaches is important to assess the significance of new approaches to self-organizing distributed systems. Compared to other fields, self-organizing real-world systems are lacking clearly defined benchmark scenarios, performance metrics, quantitative measures, and standard tools.
? Experiment protocols: Benchmarking of resilience, robustness, and scalability requires clearly defined experiment protocols and scenarios. We encourage contributions that define benchmarking standards, protocols to test resilience, and that implement benchmarking tool chains.
The design of decentralized, distributed embodied systems benefits from self-organization (e.g., resilience, scalability, and adaptivity to dynamic environments). However, most studies in self-organization on physical platforms (e.g., swarm robotics) were performed within lab environments. The real world comes with severe requirements, calling for more general methodologies, design method standardization, and validation/comparability via benchmarking tool-sets. With this Research Topic, we want to collect, benchmark, and survey novel approaches to push self-organization towards real-world applications, focusing on embodied artificial systems, such as multi-robot, cyber-physical, and socio-technical systems.
We expect research papers on how we can kick our robots out of the lab, pushing towards a novel ‘field swarm robotics’; other relevant fields are self-organizing cyber-physical systems in the wild and distributed systems for radically novel applications, and theoretical or practical studies on novel methods or applications for self-organization in the real world. We also seek for survey and perspective papers on requirements, methodology, and applications for real-world systems. We want to highlight the following requirements, methods, and benchmarking aspects.
Requirements
? Resilience: studies on decentralized systems too often take robustness for granted because of redundancy. Redundancy, however, is only a precondition for robustness. We require novel approaches that make embodied systems robust against random faults and resilient to intentional Byzantine agents. For example, revisit classical coordination algorithms and check how to (1) break them, (2) fix them, and (3) create formal quantitative measures.
? Scalability: Scalability poses stronger requirements in open systems where agents join and leave at runtime. Robots may break or be added to the system, requiring to dynamically scale and reconfigure to the system size.
? Adaptivity: Biological self-organizing systems are known to be adaptive. However, in artificial systems, it is an under-researched requirement. Real-world systems are exposed to unanticipated situations with unavoidable uncertainty, requiring online adaptations. The self-organizing embodied system may be heterogeneous and has to deal with tasks and environments of high complexity.
Methods
? Methods for radically new applications that are driven by real-world requirements
? Learning: We require novel methods of offline/online learning for increased adaptivity in self-organizing systems in open environments. We encourage contributions of online distributed (evolutionary) learning, including embodied evolutionary robotics. We also welcome the application of deep reinforcement learning to distributed embodied systems that resolve the challenge of sparse data. Contributions should address the reality gap and the challenge of designing resource-efficient learning methods.
? Design, modelling, deployment: A general methodology is required for the design of collective systems. Key design challenges are: defining the initial requirements of hardware/software, designing interactions among the cyber-physical systems, and validating the deployed system.
Benchmarking
? Benchmarking standards: Comparability of approaches is important to assess the significance of new approaches to self-organizing distributed systems. Compared to other fields, self-organizing real-world systems are lacking clearly defined benchmark scenarios, performance metrics, quantitative measures, and standard tools.
? Experiment protocols: Benchmarking of resilience, robustness, and scalability requires clearly defined experiment protocols and scenarios. We encourage contributions that define benchmarking standards, protocols to test resilience, and that implement benchmarking tool chains.
