Robust falling-down avoidance control for acrobat robot using switching controller

In this paper, we propose methods of designing systematic controller including identification to stabilize the acrobat robot with reinforcement learning algorithm. We model the final stage of the robot at falling-down as a switched system with few contact conditions. For this model, we can systematically design a SDOF controller with for an unstable LTI system whose parameters are calculated by a closed loop identification. Considering the change of the dimension of the subsystem, we add a new subsystem with a reduced dimension to the switched system. we confirm the effectiveness of these methods with experiments using a real robot system.