Direct-reinforcement-adaptive-learning neural network control for nonlinear systems

The paper is concerned with the application of reinforcement learning techniques to feedback control of nonlinear systems using neural networks (NN). Even if a good model of the nonlinear system is known, it is often difficult to formulate a control law. The work in this paper addresses this problem by showing how a NN can cope with nonlinearities through reinforcement learning with no preliminary off-line learning phase required. The learning is performed on-line based on a binary reinforcement signal from a critic without knowing the nonlinearity appearing in the system, whence comes the term ''Direct Reinforcement Adaptive Learning (DRAL).'' The DRAL algorithm is derived from Lyapunov stability analysis, so that both system tracking stability and error convergence can be guaranteed in the closed-loop system.