Abstract

Reinforcement Learning (RL) methods enable autonomous robots to learn skills from scratch by interacting with the environment. However, reinforcement learning can be very time consuming. This paper focuses on accelerating the reinforcement learning process on a mobile robot in an unknown environment. The presented algorithm is based on approximate policy iteration with a continuous state space and a fixed number of actions. The action-value function is represented by a weighted combination of basis functions.
Furthermore, a complexity analysis is provided to show that the implemented approach is guaranteed to converge on an optimal policy with less computational time.
A parallel parking task is selected for testing purposes. In the experiments, the efficiency of the proposed approach is demonstrated and analyzed through a set of simulated and real robot experiments, with comparison drawn from two well known algorithms (Dyna-Q and Q-learning).