Research map / Reinforcement learning / Visual and representation RL
Pre-trained Visual Dynamics Representations for Efficient Policy Learning
Hao Luo, Bohan Zhou, Zongqing Lu
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 longer horizon (2 context + 6 future frames) instead of 1+1 (ablation) | PVDR · Meta-World Reach | success rate (%): 7.8 -> 45.0 | Improved |
| use longer horizon (2+6) instead of 2+2 (ablation) | PVDR · Meta-World Plate Slide | success rate (%): 11.4 -> 35.2 | Improved |
| add context representation (ContextWM-style) to video prediction (ablation) | PVDR · Meta-World Reach | success rate (%): 45.0 -> 31.0 | Hurt |
| add context representation (ContextWM-style) to video prediction (ablation) | PVDR · Meta-World Plate Slide | success rate (%): 35.2 -> 22.0 | Hurt |
| pre-train on in-domain Meta-World videos vs BAIR videos (ablation) | PVDR · Meta-World Reach | success rate (%): 45.0 -> 65.4 | Improved |
| pre-train on Berkeley Cable Routing (very different) instead of BAIR (ablation) | PVDR · Meta-World Reach | success rate (%): 45.0 -> 39.0 | Hurt |
| replace selected visual dynamics representation with random tensor (ablation) | PVDR · 4 Meta-World tasks | success rate | Hurt |
| add goal-oriented term to prior loss, supervised action loss, and PPO loss (ablation) | PVDR online adaptation · 4 Meta-World tasks | success rate | Improved |
| vary loss weight lambda_1 (ablation) | PVDR · 4 Meta-World tasks | success rate | Mixed |
| use PVDR instead of APV/FICC/STG baselines | Meta-World 12 tasks · Meta-World (12 tasks) | success rate | Improved |
Closest papers by meaning
- The Surprising Ineffectiveness of Pre-Trained Visual Representations for Model-Based Reinforcement Learning (NeurIPS 2024 · similarity 0.90)
- Video Prediction Policy: A Generalist Robot Policy with Predictive Visual Representations (ICML 2025 · similarity 0.90)
- Towards Principled Representation Learning from Videos for Reinforcement Learning (ICLR 2024 · similarity 0.89)
- Learning an Actionable Discrete Diffusion Policy via Large-Scale Actionless Video Pre-Training (NeurIPS 2024 · similarity 0.89)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.88)
- Scaling Proprioceptive-Visual Learning with Heterogeneous Pre-trained Transformers (NeurIPS 2024 · similarity 0.88)
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.
