Abstract: Aiming at the stable self-balancing cognition problems of flexible self-balancing robot, a balance cognition method based on multi-level heuristic dynamic programming is proposed and applied on the self-balance learning of flexible self-balancing robot in this paper. In the proposed cognition method, the original reward mechanism with discrete form is transformed into a continuous form by introducing the orientational reward module, and the converted continuous reward signal is used as the major basis for evaluation. The scheme enables the robot to record more information in the autonomic cognition process and improve its cognitive ability. Through the robot self-balancing cognitive experiment, it can be seen that the robot can still be able to achieve good cognitive ability even the robot contains flexible joints. Its learning effect and robustness are better than traditional method.

Key words: HDP (heuristic dynamic programming), flexible self-balancing robot, cognitive model, internal reward