Long-Term Exploration in Persistent MDPs
This addresses the challenge of efficient exploration for reinforcement learning agents in complex, sparse-reward environments, representing an incremental improvement over existing curiosity-based methods.
The paper tackles the problem of exploration in hard-exploration reinforcement learning environments with sparse rewards by proposing Rollback-Explore (RbExplore), a method that allows agents to roll back to visited states in persistent MDPs, and shows that it outperforms or matches state-of-the-art curiosity methods like ICM and RND in the Prince of Persia game without rewards or domain knowledge.
Exploration is an essential part of reinforcement learning, which restricts the quality of learned policy. Hard-exploration environments are defined by huge state space and sparse rewards. In such conditions, an exhaustive exploration of the environment is often impossible, and the successful training of an agent requires a lot of interaction steps. In this paper, we propose an exploration method called Rollback-Explore (RbExplore), which utilizes the concept of the persistent Markov decision process, in which agents during training can roll back to visited states. We test our algorithm in the hard-exploration Prince of Persia game, without rewards and domain knowledge. At all used levels of the game, our agent outperforms or shows comparable results with state-of-the-art curiosity methods with knowledge-based intrinsic motivation: ICM and RND. An implementation of RbExplore can be found at https://github.com/cds-mipt/RbExplore.