Learning Markov State Abstractions for Deep Reinforcement Learning
This addresses the challenge of preserving the Markov property in deep reinforcement learning for agents with rich observations, offering a practical solution with broad applicability, though it builds incrementally on existing methods.
The paper tackles the problem of learning Markov state abstractions in reinforcement learning with rich observations, introducing a novel training procedure that combines inverse model estimation and temporal contrastive learning to improve sample efficiency. The result shows that the approach often matches or exceeds performance with hand-designed compact state information on visual gridworld and continuous control benchmarks.
A fundamental assumption of reinforcement learning in Markov decision processes (MDPs) is that the relevant decision process is, in fact, Markov. However, when MDPs have rich observations, agents typically learn by way of an abstract state representation, and such representations are not guaranteed to preserve the Markov property. We introduce a novel set of conditions and prove that they are sufficient for learning a Markov abstract state representation. We then describe a practical training procedure that combines inverse model estimation and temporal contrastive learning to learn an abstraction that approximately satisfies these conditions. Our novel training objective is compatible with both online and offline training: it does not require a reward signal, but agents can capitalize on reward information when available. We empirically evaluate our approach on a visual gridworld domain and a set of continuous control benchmarks. Our approach learns representations that capture the underlying structure of the domain and lead to improved sample efficiency over state-of-the-art deep reinforcement learning with visual features -- often matching or exceeding the performance achieved with hand-designed compact state information.