Using Interaction-Based Learning to Construct an Adaptive and Fault-Tolerant Multi-Link Floating Robot

How to build distributed autonomous systems that can adaptively behave through learning in the real world is still an open problem in the research field. In order to tackle this problem, we constructed a distributed autonomous floating robot that consisted of mechanically linked multiple identical units and proposed a new control framework, adaptive oscillator method, to deal with units’ temporal and spatial interaction with their environment. A single model reinforcement learning system was first employed to realize the framework, and a multiple-model reinforcement learning system was proposed further and employed to cope with environmental changes caused by adding obstacles. In order to confirm adaptive behavior acquisition and fault-tolerance in a real environment, we did experiments on target approaching task by using the real floating robot.