Research map / Reinforcement learning / Visual and representation RL
Escaping the Diversity Trap in Robotic Manipulation via Anchor-Centric Adaptation
Yanzhe Chen, Kevin Yuchen Ma, Qi Lv, Yiqi Lin, Zechen Bai, Chen Gao, Mike Zheng Shou
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 |
|---|---|---|---|
| add Anchor-Centric Adaptation (repeated anchors + mined boundary + residual) to standard diverse fine-tuning | π0.5 · Real-robot 4 tasks, N=100 | mean success rate (%): 31.7 -> 72.5 | Improved |
| add ACA at N=50 versus baseline at N=150 | π0.5 · Real-robot 4 tasks | mean success rate (%): 46.3 vs 52.9 | Improved |
| use error-driven boundary mining instead of random expansion (with full FT) (ablation) | ACA · Real-robot, N=100 | mean success rate (%): 57.1 -> 64.2 | Improved |
| use residual (LoRA) update instead of full fine-tuning, with mined expansion (ablation) | ACA · Real-robot, N=100 | mean success rate (%): 64.2 -> 72.5 | Improved |
| use residual update with random expansion instead of full FT (ablation) | ACA · Real-robot, N=100 | mean success rate (%): 57.1 -> 58.8 | Improved |
| add Stage 2 with random expansion and full fine-tuning on top of Stage-1 policy (ablation) | ACA · Real-robot, N=100 | mean success rate (%): 62.1 -> 57.1 | Hurt |
| increase number of distinct anchors beyond the optimum (ablation) | ACA · Real-robot, budgets 50/100/150 | success rate | Hurt |
| reduce anchor budget NA from 80 to 50 (ablation) | ACA · Real-robot, N=100 | S@1 success rate (%): 80.0 -> 65.0 | Hurt |
| raise NA from 80 to 100 (fewer Stage 2 trajectories) (ablation) | ACA · Real-robot, N=100 | S@3 / mean success: S@1 82.5; S@3 55.0 vs 65.0 | Mixed |
| use random anchor placement instead of structured Center Rect layout (ablation) | ACA · Real-robot, N=100, K=6 | mean success rate (%): 72.5 -> 70.4 / 68.8 | No clear effect |
| use spatially biased top-left anchors instead of no anchors (ablation) | ACA · Real-robot, N=100 | S@3 success rate (%): 12.5 -> 51.0 | Improved |
| use rectangular vs circular anchor arrangement (ablation) | ACA · Real-robot, N=100 | region success rates | No clear effect |
Closest papers by meaning
- PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- Unified Noise Steering for Efficient Human-Guided VLA Adaptation (NeurIPS 2026 · similarity 0.88)
- Exploring the Limits of Vision-Language-Action Manipulation in Cross-task Generalization (NeurIPS 2025 · similarity 0.88)
- Efficient Test-Time Adaptation For Robot Policies (NeurIPS 2026 · similarity 0.88)
- Solving New Tasks by Adapting Internet Video Knowledge (ICLR 2025 · similarity 0.87)
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.
