Y. Chen, Y. Li and D.J. Braun, Learning Finite-Horizon Nonlinear Optimal Control Policies with Unknown Control-Affine Dynamics (under review).
This paper introduces a model-free method for learning finite-horizon optimal control for systems with control-affine dynamics. We approximate the time- and state-dependent optimal control policy using model-free re-learning of simple linear control policies. We prove the convergence and optimality of this learning method and demonstrate its use through a numerical example.