State Indexed Policy Search by Dynamic Programming

We consider the reinforcement learning problem of simultaneous trajectory-following and obstacle avoidance by a radio-controlled car. A space-indexed non-stationary controller policy class is chosen that is linear in the features set, where the multiplier of each feature in each controller is learned using the policy search by dynamic programming algorithm. Under the control of the learned policies the radio-controlled car is shown to be capable of reliably following a predefined trajectory while avoiding point obstacles along its way.