Research map / Reinforcement learning / Visual and representation RL
RL-ViGen: A Reinforcement Learning Benchmark for Visual Generalization
Zhecheng Yuan, Sizhe Yang, Pu Hua, Can Chang, Kaizhe Hu, Huazhe Xu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use ImageNet pre-trained ResNet encoder (PIE-G) for visual generalization | PIE-G · Habitat ImageNav appearance/lighting | success rate | Improved |
| segment agent via saliency to remove background (SGQN) | SGQN · Habitat and CARLA (object-rich, first-person) | generalization score | Hurt |
| frequency-domain data augmentation (SRM) | SRM · CARLA autonomous driving | aggregated return | Improved |
| ImageNet pre-trained encoder under low-frequency scenes (dark night) | PIE-G · CARLA autonomous driving | aggregated return | Hurt |
| strong augmentation versus random shift only, with VRL3 safe Q mechanism | VRL3-based agents · Adroit sample efficiency | success rate | No clear effect |
| ImageNet pre-trained model for fine-grained features (PIE-G) | PIE-G · Adroit hard visual generalization | success rate | Improved |
| saliency-map based generalization (SGQN) | SGQN · Adroit camera views, hard | success rate | Improved |
| generalization methods for scene structure changes | all algorithms · CARLA scene structures | return | No clear effect |
| generalization-based methods versus sample-efficiency-focused methods | cross-embodiment transfer (Panda to KUKA IIWA, Kinova3) · Robosuite cross embodiment | return | Mixed |
| dynamic video backgrounds in hard setting | SRM · Robosuite table-top hard | return | Hurt |
| single shared encoder without target encoder or momentum update in CURL | CURL re-implementation · DMControl sample efficiency | episode return | Improved |
| add pre-trained encoder inference or saliency maps | PIE-G and SGQN · DMControl Walker walk | FPS | Hurt |
Closest papers by meaning
- The Surprising Ineffectiveness of Pre-Trained Visual Representations for Model-Based Reinforcement Learning (NeurIPS 2024 · similarity 0.92)
- What Can RL Bring to VLA Generalization? An Empirical Study (NeurIPS 2025 · similarity 0.90)
- Focus On What Matters: Separated Models For Visual-Based RL Generalization (NeurIPS 2024 · similarity 0.90)
- Sliding Puzzles Gym: A Scalable Benchmark for State Representation in Visual Reinforcement Learning (ICML 2025 · similarity 0.90)
- A Simple Framework for Generalization in Visual RL under Dynamic Scene Perturbations (NeurIPS 2024 · similarity 0.90)
- DrM: Mastering Visual Reinforcement Learning through Dormant Ratio Minimization (ICLR 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
