Do Agents Dream of Electric Sheep?: Improving Generalization in Reinforcement Learning through Generative Learning
This work addresses generalization issues in reinforcement learning for agents in sparse-reward settings, presenting an incremental improvement over existing imagination-based approaches.
The paper tackles the problem of poor generalization in reinforcement learning agents by using generative augmentations to modify predicted trajectories, achieving higher generalization in sparsely rewarded ProcGen environments compared to classic imagination and offline training methods.
The Overfitted Brain hypothesis suggests dreams happen to allow generalization in the human brain. Here, we ask if the same is true for reinforcement learning agents as well. Given limited experience in a real environment, we use imagination-based reinforcement learning to train a policy on dream-like episodes, where non-imaginative, predicted trajectories are modified through generative augmentations. Experiments on four ProcGen environments show that, compared to classic imagination and offline training on collected experience, our method can reach a higher level of generalization when dealing with sparsely rewarded environments.