Research map / Vision-language and multimodal / Robot action and navigation
VLA-Adapter: An Effective Paradigm for Tiny-Scale Vision-Language-Action Model
Yihao Wang, Pengxiang Ding, Lingxiao Li, Can Cui, Zirui Ge, Xinyang Tong, Wenxuan Song, Han Zhao, Wei Zhao, Pengxu Hou, Siteng Huang, Yifan Tang and 4 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 |
|---|---|---|---|
| use VLA-Adapter bridge instead of OpenVLA-OFT bridge | Prismatic Qwen2.5-0.5B (no robotic pretraining) · LIBERO-Long | success rate (%): 85.8 -> 95.0 | Improved |
| use VLA-Adapter bridge instead of OpenVLA-OFT bridge | Prismatic LLaMA2-7B · LIBERO-Long | success rate (%): 87.5 -> 95.2 | Improved |
| use VLA-Adapter bridge instead of OpenVLA-OFT bridge | OpenVLA-7B (robot-pretrained) · LIBERO-Long | success rate (%): 94.5 -> 95.4 | Improved |
| freeze the VLM backbone with OpenVLA-OFT bridge | OpenVLA-OFT · LIBERO-Long | success rate (%): 0.0 | Hurt |
| use middle-layer raw features instead of deep-layer raw features (ablation) | VLA-Adapter policy · LIBERO-Long | success rate | Improved |
| use all-layer features instead of a single layer (ablation) | VLA-Adapter policy · LIBERO-Long | success rate | Improved |
| add ActionQuery features alongside all-layer raw features (ablation) | VLA-Adapter · LIBERO-Long | success rate (%): 90.6 -> 95.0 | Improved |
| add raw features to all-layer ActionQuery condition (ablation) | VLA-Adapter · LIBERO-Long | success rate (%): 92.6 -> 95.0 | Improved |
| mid-layer raw features vs all-layer ActionQuery on hard subtasks (ablation) | VLA-Adapter · LIBERO-Long subtask 7 | success rate (%): 90 (raw layer 9) vs 82 (AQ layer 1) | Mixed |
| use learnable tanh(g) gate for raw feature injection instead of fixed 1 (ablation) | VLA-Adapter Bridge Attention · LIBERO-Long | success rate (%): 91.4 -> 95.0 | Improved |
| also gate ActionQuery injection with tanh(g) (ablation) | VLA-Adapter Bridge Attention · LIBERO-Long | success rate (%): 95.0 -> 92.6 | Hurt |
| use very few or very many ActionQuery tokens vs 64 (ablation) | VLA-Adapter · LIBERO-Long | success rate | Hurt |
Later papers that compared with this method
9 later accepted papers report VLA-Adapter as a baseline in their results tables; 8 of them report a better number on most of the results they share with it.
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · better on 11 of 11 shared results)
- StableVLA: Towards Robust Vision-Language-Action Models without Extra Data (ICML 2026 · better on 7 of 7 shared results)
- World–Value–Action Model: Implicit Planning for Vision–Language–Action Systems (NeurIPS 2026 · better on 4 of 5 shared results)
- Think Less, Act Early: Reinforced Latent Reasoning with Early Exit in Vision-Language-Action Models (ICML 2026 · better on 3 of 3 shared results)
- Spatial Representation Distillation and Knowledge Routing for Vision-Language-Action Models (NeurIPS 2026 · better on 1 of 2 shared results)
- Localizing, Structuring, and Rendering: Bridging 3D and 2D Vision-Language-Action Models for Robotic Manipulation (CVPR 2026 · better on 2 of 2 shared results)
- ACoT-VLA: Action Chain-of-Thought for Vision-Language-Action Models (CVPR 2026 · better on 2 of 2 shared results)
- ActionUNet: Improving Robustness of VLA Models with Efficient Multi-scale Fine-tuning (NeurIPS 2026 · better on 1 of 1 shared result)
1 more not listed.
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.
- Seer-Large — SEER: Backdoor Detection for Vision-Language Models through Searching Target Text and Image Trigger Jointly (AAAI 2024) · 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
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.94)
- Beyond Data Scaling: Representation-Centric Pre-training for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.94)
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.93)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.93)
- VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching (NeurIPS 2025 · similarity 0.93)
- EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models (NeurIPS 2025 · similarity 0.93)
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.
