Research map / Vision-language and multimodal / Robot action and navigation
Scaling by Diversified Experience for Vision-Language-Action Models
Leiyu Wang, Zhaofengnian Wang, Xueqi Li, Luoyi Fan, Cewu Lu, Nanyang Ye
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 Intention Decoupling (ablation) | SyVLA · Folding Shirt | average success rate: 0.43 -> 0.86 | Improved |
| add chain-of-thought reasoning together with Intention Decoupling (ablation) | SyVLA · Folding Shirt | average success rate: 0.79 -> 0.86 | Improved |
| add the RL stage (ablation) | SyVLA · Folding Shirt | average success rate: 0.71 -> 0.86 | Improved |
| add expert dataset samples during RL (ablation) | SyVLA · Folding Shirt | average success rate: 0.21 -> 0.86 | Improved |
| retrieve similar expert samples instead of random ones (ablation) | SyVLA · Folding Shirt | average success rate: 0.79 -> 0.86 | Improved |
| retrieve similar expert samples instead of random ones (ablation) | SyVLA · Real-world Task 2 and Task 3 | success rate: 0.56 -> 0.68 (Task 2); 0.571 -> 0.643 (Task 3) | Improved |
| use constant advantage 1.0 for expert samples instead of standard GAE (ablation) | SyVLA RL · Folding Shirt | average success rate: 0.00 -> 0.86 | Improved |
| raise mask threshold from 5% to 20% (ablation) | Intention Decoupling · Folding Shirt | average success rate: 0.86 -> 0.64 | Hurt |
| raise mask threshold from 3% to 5% (ablation) | Intention Decoupling · Folding Shirt | average success rate: 0.73 -> 0.86 | Improved |
| increase Feature Query Tokens from 5 to 20 (ablation) | SyVLA · Folding Shirt | average success rate: 0.39 -> 0.86 | Improved |
| increase Feature Query Tokens from 20 to 30 (ablation) | SyVLA · Folding Shirt | average success rate: 0.86 -> 0.87 | No clear effect |
| add Intention Decoupling (ablation) | SyVLA · multimodal benchmarks (DocVQA, AI2D, MMMU, MME, HallBench) | score: DocVQA 63.62 -> 80.01; AI2D 71.76 -> 67.7; MMMU 38.67 -> 35.78; MME 1648.12 -> 1795; HallBench 41.7 -> 42.53 | Mixed |
Closest papers by meaning
- VLA-R1: Enhancing Reasoning in Vision-Language-Action Models (ECCV 2026 · similarity 0.93)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.93)
- VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching (NeurIPS 2025 · similarity 0.91)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- VLA-Corrector: Lightweight Detect-and-Correct Inference for Adaptive Action Horizon (NeurIPS 2026 · similarity 0.90)
- PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models (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.
