Research map / Vision-language and multimodal / Robot action and navigation
MA-VLA: Multi-Arm Vision-Language-Action Model for Collaboration and Compositional Generalization
Zaibin Zhang, Junlan Xiao, Zhongbo Zhang, Yifan Wang, Li Kang, Yiran Qin, Changxing Xia, Heng Zhou, Talas Fu, Enshen Zhou, Ruimao Zhang, Zhenfei Yin and 2 more
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 atomic action prompts (ablation) | Pi0 · Three Robots Stack Cube (in-domain) | success rate: 48.0% -> 58.0% | Improved |
| add atomic action prompts (ablation) | Pi0 · Three Robots Stack Cube (out-of-domain) | success rate: 0.0% -> 0.0% | No clear effect |
| add Arm Shuffle (ablation) | MA-VLA · Three Robots Stack Cube (out-of-domain) | success rate: 0.0% -> 7.3% | Improved |
| add Arm Shuffle (ablation) | MA-VLA · Three Robots Stack Cube (in-domain) | success rate: 58.0% -> 52.0% | Hurt |
| add View Dropout (ablation) | MA-VLA · Three Robots Stack Cube (out-of-domain) | success rate: 7.3% -> 15.3% | Improved |
| add View Dropout (ablation) | MA-VLA · Three Robots Stack Cube (in-domain) | success rate: 52.0% -> 53.0% | No clear effect |
| use one unified VLA instead of three separate per-arm Pi0 models (ablation) | MA-VLA · Three Robots Stack Cube (out-of-domain) | success rate: 0.0% -> 15.3% | Improved |
| use one unified VLA instead of three separate per-arm Pi0 models (ablation) | MA-VLA · Three Robots Stack Cube (in-domain) | success rate: 61.0% -> 53.0% | Hurt |
| raise Arm Shuffle probability from 0% to 100% (ablation) | MA-VLA · Out-of-domain Stack Cube | success rate | Improved |
| use MA-VLA instead of Pi0 backbone on RoboFactory two-arm | Pi0 · RoboFactory two-arm tasks (avg) | success rate: 80.3% -> 83.5% | Improved |
| use MA-VLA instead of Pi0 on three- and four-arm tasks | Pi0 · RoboFactory three- and four-arm tasks (avg) | success rate: 76.5% -> 83.3% | Improved |
| use MA-VLA instead of Pi0 on RoboTwin 2.0 Hard | Pi0 · RoboTwin 2.0 Hard (avg) | success rate: 41.1 -> 49.0 | Improved |
Closest papers by meaning
- Beyond World-Frame Action Heads: Motion-Centric Action Frames for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.91)
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- Exploring the Limits of Vision-Language-Action Manipulation in Cross-task Generalization (NeurIPS 2025 · similarity 0.90)
- VLA-ATTC: Adaptive Test-Time Compute for VLA Models with Relative Action Critic Model (ICML 2026 · similarity 0.90)
- VLA-OS: Structuring and Dissecting Planning Representations and Paradigms in Vision-Language-Action Models (NeurIPS 2025 · similarity 0.90)
- ActionUNet: Improving Robustness of VLA Models with Efficient Multi-scale Fine-tuning (NeurIPS 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.
