Mobile robot navigation based on improved CA-CMAC and Q-learning in dynamic environment

An algorithm based on Credit Assigned Cerebellar Model Articulation Controller(CA-CMAC) fitting Q-learning and improved by a learning parameter of adaptive adjusting balance, is proposed in this paper, which is applied for solving the mobile robot navigation problem in complex environment. Compared to the traditional navigation algorithm, this algorithm avoids local minimum problem, and has a better adaptability and real-time policy-making ability. The local approximation characteristic of the cerebellum model is used to overcome the problem of conventional universal neural network, such as BP, which results in slow convergence speed of neural network or cannot converge. Furthermore, the convergence speed of this algorithm is further accelerated due to introduction of the adaptive adjustment balance parameter. Simulation results indicate that the proposed algorithm improves the algorithm performance, at the same time, whether it is a path, in the simple static obstacles environment or in the complex dynamic obstacles environment, can be successful planned by mobile robot.