Research on reinforcement learning and its application to mobile robot

Abstract

Path planning,both global and local, is one of key problems of intelligent robot. Local path planning is particularly challenging. It is difficult to obtain a good path in a complex environment. To overcome this,the reinforcement learning method was used for local path planning in a unknown and complex environment. Aiming at the slow convergent rate and other drawbacks of standard Q-learning, the multi-step on-policy SARSA(λ) reinforcement algorithm was adopted in the field of robot's local path planning and the related problems were discussed. The CMAC neural network was used to realize the reinforcement learning system. The SARSA(λ) algorithm was implemented by using the CMAC neural network. The switching method between the path planning network and the wall-following network was adopted to resolve local path planning of a autonomous mobile robot in the environment with complex obstacles. The simulation results demonstrate the efficiency of the algorithm. The algorithm has good (adaptation) and self-learning ability compare with other traditional methods.