Research map / Reinforcement learning / Visual and representation RL
Oracle-Guided Masked Contrastive Reinforcement Learning for Visuomotor Policies
Yuhang Zhang, Jiaping Xiao, Chao Yan, Mir Feroskhan
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 linear decay instead of fixed oracle KL coefficient (ablation) | OMC-RL · Tree (Hard) | success rate (%): 71.0 -> 78.5 | Improved |
| use linear decay instead of exponential decay (ablation) | OMC-RL · Tree (Hard) | success rate (%): 72.5 -> 78.5 | Improved |
| use linear decay instead of fixed coefficient (ablation) | OMC-RL · Furniture (Easy) | success rate (%): 81.5 -> 91.5 | Improved |
| use masked temporal contrastive learning instead of consecutive frames in CURL (ablation) | CURL · Furniture (Easy) | success rate (%): 30.5 -> 72.5 | Improved |
| feed consecutive frames to CURL (ablation) | CURL · Barrier (Medium) | success rate (%): 30.0 -> 0.0 | Hurt |
| add masked contrastive plus oracle guidance vs CURL | OMC-RL · Barrier (Medium) | success rate (%): 30.0 -> 84.5 | Improved |
| add oracle teacher (ablation) | OMC-RL · Barrier and Tree scenes | navigation metrics | Improved |
| add oracle teacher (ablation) | OMC-RL · Furniture (Easy) | navigation metrics | Mixed |
| add oracle teacher to NPE (ablation) | NPE · simulation | navigation metrics | Improved |
| add non-linear projection head (ablation) | OMC-RL · simulation | navigation metrics | Improved |
| add non-linear projection head (ablation) | CURL · simulation | navigation metrics | Improved |
| add Transformer to key encoder as well (dual) (ablation) | OMC-RL · simulation | navigation metrics | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CURL — CURL: Contrastive Unsupervised Representations for Reinforcement Learning (ICML 2020) · better on 7 of 7 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Latent Policy Barrier: Learning Robust Visuomotor Policies by Staying In-Distribution (NeurIPS 2025 · similarity 0.89)
- DMC-VB: A Benchmark for Representation Learning for Control with Visual Distractors (NeurIPS 2024 · similarity 0.87)
- DynaMo: In-Domain Dynamics Pretraining for Visuo-Motor Control (NeurIPS 2024 · similarity 0.87)
- Falcon: Fast Visuomotor Policies via Partial Denoising (ICML 2025 · similarity 0.86)
- FreqPolicy: Efficient Flow-based Visuomotor Policy via Frequency Consistency (NeurIPS 2025 · similarity 0.86)
- FreqPolicy: Frequency Autoregressive Visuomotor Policy with Continuous Tokens (NeurIPS 2025 · similarity 0.86)
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.
