Research map / Vision-language and multimodal / Robot action and navigation
Less is More: Reducing Complexity in Vision-Language-Action Systems
Jinhui Ye, Ning Gao, Senqiao Yang, Jinliang Zheng, Zixuan Wang, Yuxin Chen, Pengguang Chen, Yilun Chen, Shu Liu, Jiaya Jia
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 continuous MLP action regression instead of discrete FAST tokens (ablation) | StarVLA-α · SimplerEnv WidowX | success rate: 35.6 -> 64.6 | Improved |
| use continuous MLP action regression instead of discrete FAST tokens (ablation) | StarVLA-α · RoboCasa-GR1 | success rate: 45.0 -> 53.8 | Improved |
| replace MLP head with GR00T-style dual-system flow-matching head (ablation) | StarVLA-α · RoboCasa-GR1 | success rate: 53.8 -> 52.8 | No clear effect |
| replace MLP head with pi0-style flow-matching expert (ablation) | StarVLA-α · RoboCasa-GR1 and WidowX | success rate: 53.8 -> 48.9 RoboCasa; 64.6 -> 65.9 WidowX | Mixed |
| add OXE robot pretraining (ablation) | StarVLA-α · RoboTwin 2.0 Clean 50x50 | success rate: 50.3 -> 30.2 | Hurt |
| add OXE robot pretraining (ablation) | StarVLA-α · RoboCasa-GR1 24x1000 | success rate: 53.8 -> 27.8 | Hurt |
| add InternData-A1 pretraining (ablation) | StarVLA-α · RoboTwin 2.0 Clean 50x50 | success rate: 50.3 -> 63.6 | Improved |
| add InternData-A1 pretraining (ablation) | StarVLA-α · RoboCasa-GR1 24x1000 | success rate: 53.8 -> 35.4 | Hurt |
| add proprioceptive state input (ablation) | StarVLA-α · RoboTwin 2.0 Clean 50x50 | success rate: 50.3 -> 60.8 | Improved |
| add stacked history frames (ablation) | StarVLA-α · RoboCasa-GR1 24x100 | success rate: 39.4 -> 33.2 | Hurt |
| predict delta actions instead of absolute (ablation) | StarVLA-α · RoboCasa-GR1 24x10 | success rate: 9.8 -> 15.8 | Improved |
| use simple zero padding instead of multi-action head (ablation) | StarVLA-α generalist · RoboCasa-GR1 | success rate: 53.5 -> 57.3 | Improved |
Closest papers by meaning
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.93)
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.92)
- Beyond World-Frame Action Heads: Motion-Centric Action Frames for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.92)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.92)
- ActionUNet: Improving Robustness of VLA Models with Efficient Multi-scale Fine-tuning (NeurIPS 2026 · similarity 0.92)
- Beyond Data Scaling: Representation-Centric Pre-training for Vision-Language-Action Models (NeurIPS 2026 · 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.
