A hybrid environment adaptation controller for a snake-like robot with online and autonomous learning ability

In this paper, we propose a two-modes autonomous controller for a snake-like robot to adapt to the environments with different friction coefficients. According to the cyclic 3-stages experiential learning theory, the controller integrates an open-form searching method and a closed-form searching method, which correspond to the two modes of the controller respectively. We introduce a mechanism to remember the already derived optimal relationship, thereby gradually approximating the optimal relationship for all environments. The controller can learn online so the approximation can be improved online and the controller will be also more and more accurate. The learning is also autonomous. After enough learning, the approximation will be very accurate so the optimum can be attained just by predicting rather than learning, thereby avoiding unnecessary leaning. Therefore, the controller is efficient. The accuracy and efficiency are validated in the simulations.