Task-Relevant Representation Decoupling for Visual Reinforcement Learning Generalization
About
Visual Reinforcement Learning (VRL) has achieved considerable success in solving control tasks. However, generalizing learned policies to new environments remains a major challenge, as agents often overfit to task-irrelevant features in the training environment. To solve this problem, we introduce the concept of decoupling observations into task-relevant and task-irrelevant representations. Building on this idea, we propose a self-supervised Task-Relevant Representation Decoupling (T2RD) algorithm for VRL. This algorithm consists of three components: task-relevant representation consistency, cross-reconstruction, and cross-dynamic prediction. The first two components achieve the decoupling of content and style features, but the resulting content representations are not necessarily task-relevant. To further refine task-relevant features from content representations, we design the third component that introduces dynamic prediction. T2RD achieves State-Of-The-Art (SOTA) generalization performance and sample efficiency in the DeepMind Control Suite and Robotic Manipulation tasks.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Reinforcement Learning | DMC-GB2 Video Hard (test) | Cartpole Swingup Return725 | 15 | |
| Robotic Manipulation | peg-in-box (test2) | Return129.4 | 14 | |
| Visual Reinforcement Learning | DMC-GB Color Hard | Average Return: Walker, Walk849 | 10 | |
| Visual Reinforcement Learning | DMControl-GB Video-Easy | Walker Walk Score821 | 10 | |
| Peg in box | Robotic Manipulation (Test1) | Episode Return164.1 | 7 | |
| Peg in box | Robotic Manipulation (Test3) | Episode Return183.6 | 7 | |
| Peg in box | Robotic Manipulation (Test4) | Episode Return149.8 | 7 | |
| Reach | Robotic Manipulation (Test1) | Episode Return25.1 | 7 | |
| Reach | Robotic Manipulation (Test3) | Episode Return29.8 | 7 | |
| Reach | Robotic Manipulation (Test4) | Episode Return28.3 | 7 |