Research map / Vision-language and multimodal / Robot action and navigation
Affordance Field Intervention: Enabling VLAs to Escape Memory Traps in Robotic Manipulation
Siyu Xu, Zijian Wang, Yunke Wang, Chenghao Xia, Tao Huang, Chang 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 |
|---|---|---|---|
| add AFI (SAF-guided intervention) to pi0 | pi0 · Real-world Place Carrot (avg over 5 conditions) | success rate: 61.0% -> 87.0% | Improved |
| add AFI to pi0 | pi0 · Real-world Remove Lid (avg) | success rate: 63.0% -> 80.0% | Improved |
| add AFI to pi0.5 | pi0.5 · Real-world Stack Tape (avg) | success rate: 61.0% -> 82.0% | Improved |
| ensemble pi0 and pi0.5 candidates under SAF scorer | AFI · Real-world Stack Tape (avg) | success rate: 86.0% -> 89.0% | Improved |
| add AFI to pi0.5 on LIBERO-Pro Spatial | pi0.5 · LIBERO-Pro Spatial OOD | success rate: 54.0% -> 75.7% | Improved |
| add AFI to pi0.5 on LIBERO-Pro Object | pi0.5 · LIBERO-Pro Object OOD | success rate: 56.4% -> 73.2% | Improved |
| add rollback to high-affordance position (ablation) | pi0-AFI · Real-world position shift ablation (20 trials) | successes out of 20 trials: 8/20 -> 13/20 | Improved |
| use adaptive proprioceptive trap detection instead of fixed-step intervention (ablation) | pi0-AFI · Real-world position shift ablation (20 trials) | successes out of 20 trials: 12/20 -> 13/20 | Improved |
| increase SAF waypoints from 10 to 13 (ablation) | pi0-AFI · Real-world position shift, waypoint count (pi0-AFI) | success rate: 65.0% -> 60.0% | Hurt |
| increase SAF waypoints from 3 to 10 (ablation) | pi0-AFI · Real-world position shift, waypoint count (pi0-AFI) | success rate: 35.0% -> 65.0% | Improved |
| use AFI hybrid instead of zero-shot VLM planner ReKep | Place Carrot · Real-world Place Carrot (avg over 5 conditions) | success rate: 36.0% -> 87.0% | Improved |
Closest papers by meaning
- CoA-VLA: Improving Vision-Language-Action Models via Visual-Text Chain-of-Affordance (ICCV 2025 · similarity 0.94)
- PALM: Progress-Aware Policy Learning via Affordance Reasoning for Long-Horizon Robotic Manipulation (CVPR 2026 · similarity 0.92)
- EchoVLA: Robotic Vision-Language-Action Model with Synergistic Declarative Memory for Mobile Manipulation (ECCV 2026 · similarity 0.91)
- AffordDP: Generalizable Diffusion Policy with Transferable Affordance (CVPR 2025 · similarity 0.91)
- TraceVLA: Visual Trace Prompting Enhances Spatial-Temporal Awareness for Generalist Robotic Policies (ICLR 2025 · similarity 0.90)
- Seeing Across Views: Benchmarking Spatial Reasoning of Vision-Language Models in Robotic Scenes (ICLR 2026 · 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.
