Decoupling Value and Policy for Generalization in Reinforcement Learning
This addresses generalization issues in reinforcement learning for AI agents, particularly in visual domains, though it is incremental as it builds on existing actor-critic methods.
The paper tackles the problem of overfitting in deep reinforcement learning due to shared representations for policy and value functions by proposing IDAAC, which decouples their optimization and adds an invariance loss, achieving state-of-the-art results on the Procgen benchmark and outperforming methods on DeepMind Control tasks with distractors.
Standard deep reinforcement learning algorithms use a shared representation for the policy and value function, especially when training directly from images. However, we argue that more information is needed to accurately estimate the value function than to learn the optimal policy. Consequently, the use of a shared representation for the policy and value function can lead to overfitting. To alleviate this problem, we propose two approaches which are combined to create IDAAC: Invariant Decoupled Advantage Actor-Critic. First, IDAAC decouples the optimization of the policy and value function, using separate networks to model them. Second, it introduces an auxiliary loss which encourages the representation to be invariant to task-irrelevant properties of the environment. IDAAC shows good generalization to unseen environments, achieving a new state-of-the-art on the Procgen benchmark and outperforming popular methods on DeepMind Control tasks with distractors. Our implementation is available at https://github.com/rraileanu/idaac.