Deep RePReL--Combining Planning and Deep RL for acting in relational domains

Harsha Kokel, Arjun Manoharan, Sriraam Natarajan, Balaraman Ravindran, Prasad Tadepalli

Keywords: Planning, Abstraction, Hierarchical RL, DQN

TL;DR: Combining a symbolic planner and a Deep RL agent can help achieve the generalization ability of the planner with the effective learning ability of Deep RL

Abstract: We consider the problem of combining symbolic planning and deep reinforcement learning (RL) to achieve the best of both worlds -- the generalization ability of the planner with the effective learning ability of deep RL. To this effect, we extend a previous work of Kokel et al. 2021, RePReL, to deep RL. As we demonstrate in experiments in two relational worlds, this combined framework enables effective learning, transfer and generalization when compared to the use of an end-to-end deep RL framework.