A reinforcement learning control scheme for nonlinear systems with multiple actions

In this paper an attempt is made to apply reinforcement learning schemes to the adaptive control of nonlinear systems with multiple continuous control actions. The control task is formulated into a sequential optimization problem. A learning algorithm is developed based on the concepts of dynamic programming and stochastic approximation and the techniques of random search and parameter estimation. The proposed algorithm is complete and general enough so that the controller can be constituted by various computing models, e.g., neural networks. The efficiency of the proposed algorithm is demonstrated by applying the methods to the nonlinear control problems with multiple control actions.