Robot path planning by artificial potential field optimization based on reinforcement learning with fuzzy state

Temporal difference (TD) learning with fuzzy state is applied to robot navigation in a multi-obstacle environment. An interpretation of the state evaluation function is given by regarding the state evaluation as a discrete artificial potential field (APF). Global optimal path planning is implemented with the APF obtained by TD learning. The APF obtained is globally optimal and avoids the local minimum areas, which always appear in traditional APF methods. Fuzzy state is introduced to improve the learning efficiency. A computer evaluation experiment shows the method's effectiveness and efficiency.