Exploiting the Sign of the Advantage Function to Learn Deterministic Policies in Continuous Domains
This work addresses a theoretical gap for researchers in reinforcement learning, offering incremental improvements by extending an existing approach with a new algorithm.
The paper tackles the lack of theoretical justification for a policy update method used in deterministic policy learning in continuous domains, providing a theoretical explanation and extending it into a new trust region algorithm called Penalized NFAC (PeNFAC). The result is that PeNFAC surpasses state-of-the-art algorithms in several classic control problems, as demonstrated experimentally.
In the context of learning deterministic policies in continuous domains, we revisit an approach, which was first proposed in Continuous Actor Critic Learning Automaton (CACLA) and later extended in Neural Fitted Actor Critic (NFAC). This approach is based on a policy update different from that of deterministic policy gradient (DPG). Previous work has observed its excellent performance empirically, but a theoretical justification is lacking. To fill this gap, we provide a theoretical explanation to motivate this unorthodox policy update by relating it to another update and making explicit the objective function of the latter. We furthermore discuss in depth the properties of these updates to get a deeper understanding of the overall approach. In addition, we extend it and propose a new trust region algorithm, Penalized NFAC (PeNFAC). Finally, we experimentally demonstrate in several classic control problems that it surpasses the state-of-the-art algorithms to learn deterministic policies.