Environmental adaptation for wheel-legged robot using external force given to legs

In recent years, autonomous mobile robots working in unknown environments are required to select apposite actions depending on their environments by themselves. In this research, we develop a system for a wheel-legged robot which grasps the situation of surrounding environment from the transition of its internal sensor state and selects the action to escape from the situation where it cannot move by itself. In our previous research, the effectiveness of the proposed system was demonstrated by the experiments with a simulator using physics engine. However, it was not shown whether the proposed system is also effective in actual environments. In this paper, we show the effectiveness of the proposed system by the experiments with a real robot and online learning in actual environments. The results of the series of experiments show that the system converges into appropriate actions in sufficiently short learning period, which means that the proposed system is effective in the actual environment.