Research map / Vision-language and multimodal / Robot action and navigation
Vlaser: Vision-Language-Action Model with Synergistic Embodied Reasoning
Ganlin Yang, Tianyi Zhang, Haoran Hao, Weiyun Wang, Yibin Liu, Dehui Wang, Guanzhou Chen, Zijian Cai, Junting Chen, Weijie Su, Wengang Zhou, Yu Qiao and 6 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 |
|---|---|---|---|
| fine-tune on Vlaser-6M embodied reasoning data | InternVL3-2B · 12 embodied reasoning benchmarks | normalized average: 15.2 -> 45.3 | Improved |
| fine-tune on Vlaser-6M embodied reasoning data | InternVL3-8B · 12 embodied reasoning benchmarks | normalized average: 22.3 -> 51.3 | Improved |
| use 8B instead of 2B model | Vlaser · point grounding vs planning and simulation | benchmark score: Where2place 74.0 (2B) vs 69.5 (8B); Ego-Plan2 38.3 (2B) vs 53.4 (8B) | Mixed |
| initialize VLA from Vlaser-OOD (out-of-domain embodied reasoning data only) instead of InternVL3-2B (ablation) | Vlaser VLA · SimplerEnv WidowX | average success rate: 41.8% -> 43.2% | No clear effect |
| initialize VLA from Vlaser-OOD instead of InternVL3-2B (ablation) | Vlaser VLA · SimplerEnv Google Robot Variant Aggregation | average success rate: 54.7% -> 51.3% | Hurt |
| initialize VLA from Vlaser-OOD instead of InternVL3-2B (ablation) | Vlaser VLA · RoboTwin | average success rate: 55.8% -> 54.5% | No clear effect |
| add in-domain QA data to VLM pretraining (ablation) | Vlaser-2B VLA · SimplerEnv WidowX | average success rate: 41.8% -> 62.6% | Improved |
| add in-domain grounding data to VLM pretraining (ablation) | Vlaser-2B VLA · SimplerEnv Google Robot Visual Matching | average success rate: 64.0% -> 73.6% | Improved |
| add in-domain spatial data to VLM pretraining (ablation) | Vlaser-2B VLA · SimplerEnv WidowX | average success rate: 41.8% -> 60.8% | Improved |
| combine QA, spatial and grounding in-domain data (Vlaser-All) vs single type (ablation) | Vlaser-2B VLA · SimplerEnv WidowX | average success rate: 62.6% (QA) -> 65.1% (All) | Improved |
| combine all in-domain data (ablation) | Vlaser-2B VLA · RoboTwin | average success rate: 55.8% -> 67.5% | Improved |
| shorten execute action length from 4 to 2 (ablation) | Vlaser-QA VLA · SimplerEnv WidowX | average success rate: 62.6% -> 51.1% | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- RoboBrain2.0-3B — RoboBrain: A Unified Brain Model for Robotic Manipulation from Abstract to Concrete (CVPR 2025) · better on 3 of 3 shared results
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
- UP-VLA: A Unified Understanding and Prediction Model for Embodied Agent (ICML 2025 · similarity 0.92)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.92)
- Knowledge Insulating Vision-Language-Action Models: Train Fast, Run Fast, Generalize Better (NeurIPS 2025 · similarity 0.91)
- Beyond Data Scaling: Representation-Centric Pre-training for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.91)
- CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models (CVPR 2025 · similarity 0.91)
- Seeing Across Views: Benchmarking Spatial Reasoning of Vision-Language Models in Robotic Scenes (ICLR 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.
