Research map / Vision-language and multimodal / Robot action and navigation
Spatial-Aware VLA Pretraining through Visual-Physical Alignment from Human Videos
Yicheng Feng, Wanpeng Zhang, Ye Wang, Hao Luo, Haoqi Yuan, Sipeng Zheng, 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 |
|---|---|---|---|
| remove spatial-aware pretraining (i.e. add it) (ablation) | VIPA-VLA · LIBERO | avg success (%): 91.2 -> 92.4 | Improved |
| add dual 3D encoder (Cut3R) fusion (ablation) | VIPA-VLA · LIBERO | avg success (%): 90.4 -> 92.4 | Improved |
| add both pretraining and dual encoder (ablation) | VIPA-VLA · LIBERO | avg success (%): 88.7 -> 92.4 | Improved |
| pretrain on Hand3D visual annotations (ablation) | InternVL3.5 · Hand3D-test | distance error (m): 0.18 -> 0.14 | Improved |
| add 3D encoder with fusion on top of Hand3D pretraining (ablation) | InternVL3.5 + Hand3D · Hand3D-test | distance error (m): 0.14 -> 0.12 | Improved |
| spatial-aware pretraining vs GR00T N1.5 on Wipe-Board | VIPA-VLA · Real robot Wipe-Board | whole-task success (%): 30% -> 60% | Improved |
| spatial-aware pretraining on Put-Three-Obj whole-task vs GR00T N1.5 | VIPA-VLA · Real robot Put-Three-Obj | whole-task success (%): 40% -> 10% | Hurt |
| VIPA-VLA vs GR00T N1.5 sub-task success | VIPA-VLA · Real robot Put-Three-Obj | sub-task success (%): 48% -> 52% | Improved |
| VIPA-VLA in unseen environments vs baselines | VIPA-VLA · Real robot Wipe-Board-Unseen | whole-task success (%): 10% -> 50% | Improved |
| human-video pretraining without robot data vs π0.5 (two-view) | VIPA-VLA · LIBERO avg | success rate (%): 96.9 -> 96.8 | No clear effect |
| VIPA-VLA single-view vs GR00T N1.5 | VIPA-VLA · LIBERO avg | success rate (%): 92.1 -> 92.4 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- CoT-VLA — CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models (CVPR 2025) · better on 1 of 1 shared result
- ThinkAct — ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planning (NeurIPS 2025) · better on 1 of 1 shared result
- 4D-VLA — 4D-VLA: Spatiotemporal Vision-Language-Action Pretraining with Cross-Scene Calibration (NeurIPS 2025) · better on 1 of 1 shared result
- TraceVLA — TraceVLA: Visual Trace Prompting Enhances Spatial-Temporal Awareness for Generalist Robotic Policies (ICLR 2025) · better on 1 of 1 shared result
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
- Beyond Data Scaling: Representation-Centric Pre-training for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.92)
- 3D-VLA: A 3D Vision-Language-Action Generative World Model (ICML 2024 · similarity 0.91)
- FineVLA: Fine-Grained Instruction Alignment for Steerable Vision-Language-Action Policies (NeurIPS 2026 · similarity 0.91)
- 4D-VLA: Spatiotemporal Vision-Language-Action Pretraining with Cross-Scene Calibration (NeurIPS 2025 · similarity 0.91)
- UP-VLA: A Unified Understanding and Prediction Model for Embodied Agent (ICML 2025 · similarity 0.91)
- BridgeVLA: Input-Output Alignment for Efficient 3D Manipulation Learning with Vision-Language Models (NeurIPS 2025 · similarity 0.91)
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.
