In this paper, we investigate the use of parallelization in reinforcement learning (RL), with the goal of learning optimal policies for single-agent RL problems more quickly by using parallel hardware. Our approach is based on agents using the SARSA(lambda) algorithm, with value functions represented using, linear function approximators. In our proposed method, each agent learns independently in a separate simulation of the single-agent problem. The agents periodically exchange information extracted from the weights of their approximators, accelerating convergence towards the optimal policy. We develop three increasingly efficient versions of this approach to parallel RL, and present empirical results for an implementation of the methods on a Beowulf cluster.
|Title of host publication||ADAPTIVE AGENTS AND MULTI-AGENT SYSTEMS|
|Editors||K Tuyls, A Nowe, Z Guessoum, D Kudenko|
|Place of Publication||BERLIN|
|Number of pages||15|
|Publication status||Published - 2008|