Learning from Visual Observation via Offline Pretrained State-to-Go Transformer
This addresses the problem of enabling reinforcement learning from video-only data for AI agents, offering a more efficient and open-ended approach compared to methods requiring additional task-specific information.
The paper tackles learning policies from visual observation data alone by proposing a two-stage framework with an offline pretrained State-to-Go Transformer, which predicts latent transitions and provides intrinsic rewards for reinforcement learning. Empirical results on Atari and Minecraft show it outperforms baselines and sometimes matches performance of policies learned from environmental rewards.
Learning from visual observation (LfVO), aiming at recovering policies from only visual observation data, is promising yet a challenging problem. Existing LfVO approaches either only adopt inefficient online learning schemes or require additional task-specific information like goal states, making them not suited for open-ended tasks. To address these issues, we propose a two-stage framework for learning from visual observation. In the first stage, we introduce and pretrain State-to-Go (STG) Transformer offline to predict and differentiate latent transitions of demonstrations. Subsequently, in the second stage, the STG Transformer provides intrinsic rewards for downstream reinforcement learning tasks where an agent learns merely from intrinsic rewards. Empirical results on Atari and Minecraft show that our proposed method outperforms baselines and in some tasks even achieves performance comparable to the policy learned from environmental rewards. These results shed light on the potential of utilizing video-only data to solve difficult visual reinforcement learning tasks rather than relying on complete offline datasets containing states, actions, and rewards. The project's website and code can be found at https://sites.google.com/view/stgtransformer.