Engineering Resilient Swarm Robotics Systems

点击翻译按钮获取中文摘要

Multi-robot systems (a.k.a. `swarms') are becoming pervasive. Although recent technological advancements demonstrate their potential in transportation, manufacturing, medical systems, and disaster response, numerous challenges must be addressed to safely realize their potential. During the 2016 Fort McMurray wildfire, for example, a swarm of robots with heterogeneous capability (e.g. flying vs. driving) could have been used to assess the scale of the disaster, locate people in danger, and provide a robust communication network for responders. Unfortunately, these functions are not possible with current technology; how to engineer collective behavior and assure safety and resilience of robot swarms, particularly in challenging and unpredictable real-world conditions, are open research problems. Robotics, artificial intelligence, and the Internet-of-Things are driving current research for the technology sector, and multi-robot systems are one of the main topics. However, current software standards for critical systems (e.g. industrial and aerospace) are not directly applicable to distributed systems which exhibit emergence - i.e. collective behaviour that is not apparent from a single robot's code. We believe that detecting and correcting issues specific to emergent systems during development should be an integral part of system design. Furthermore, we consider how swarms' behaviour and goals could be updated during missions to adapt to changes in the environment or priorities. In this research program, we propose to address these issues along 3 lines of research: Aim I) Engineer resilient collective behavior in robot swarms: we will define the theoretical foundations that allow swarms to perceive problems and adapt appropriately; Aim II) Embed safety primitives in the Buzz programming language: we will verify safety properties of collective systems during the development phase Aim III) Develop software management strategies for large robot teams: we will study novel methods for deploying and releasing software revisions to deployed robotic networks, considering their very peculiar requirements in terms of mobility and resources Overall, this research will lead to practical swarm-robotics applications. We envision disaster response (in situations such as the Fort McMurray fire) as a first demonstration of our systems, where robots are deployed and provide constant situation monitoring, assist in victim search, and provide a communication infrastructure in a fully-autonomous, self-adaptive way, with minimal human supervision.