Research map / Vision-language and multimodal / Robot action and navigation
Think Less, Act Early: Reinforced Latent Reasoning with Early Exit in Vision-Language-Action Models
Dianqiao Lei, Lianlei Shan
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 latent reasoning (ablation) | AVA-VLA · LIBERO | Average success rate (%): 95.8 -> 98.3 | Improved |
| add RL denoising (ablation) | AVA-VLA · LIBERO | Average success rate (%): 96.6 -> 98.3 | Improved |
| add latent smoothness regularization (ablation) | AVA-VLA · LIBERO | Average success rate (%): 96.9 -> 98.3 | Improved |
| add early exit (ablation) | AVA-VLA · LIBERO | Average success rate (%): 98.0 -> 98.3 | No clear effect |
| add early exit (ablation) | AVA-VLA · CALVIN ABC->D | Average length: 4.61 -> 4.65 | Improved |
| add early exit (ablation) | AVA-VLA · LIBERO-Spatial | Mean latency (ms): 312 -> 145 | Improved |
| add POMDP formulation and RL denoising to a BC latent module (ablation) | AVA-VLA · LIBERO | Average success rate (%): 96.6 -> 98.3 | Improved |
| replace truncated explicit CoT with parallel latent reasoning at matched latency (ablation) | AVA-VLA · LIBERO | Success rate (%): 78.8 -> 98.3 | Improved |
| raise fixed reasoning depth from 5 to 10 (ablation) | AVA-VLA · LIBERO | Success rate (%): 98.0 -> 98.1 | No clear effect |
| use adaptive depth instead of direct single step (ablation) | AVA-VLA · LIBERO | Success rate (%): 94.2 -> 98.3 | Improved |
| lower early-exit threshold from 0.55 to 0.30 (ablation) | AVA-VLA · LIBERO | Average success rate (%): 98.3 -> 96.1 | Hurt |
| add RL denoising (ablation) | AVA-VLA · LIBERO+ (7 perturbations) | Average success rate (%): 83.2 -> 89.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- VLA-Adapter — VLA-Adapter: An Effective Paradigm for Tiny-Scale Vision-Language-Action Model (AAAI 2026) · better on 3 of 3 shared results
- CoT-VLA — CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models (CVPR 2025) · 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
- LaST-VLA: Thinking in Latent Spatio-Temporal Space for Vision-Language-Action in Autonomous Driving (NeurIPS 2026 · similarity 0.91)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models (NeurIPS 2025 · similarity 0.90)
- More Thought, Less Accuracy? On the Dual Nature of Reasoning in Vision-Language Models (ICLR 2026 · similarity 0.90)
- ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planning (NeurIPS 2025 · similarity 0.90)
- NoisyRollout: Reinforcing Visual Reasoning with Data Augmentation (NeurIPS 2025 · 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.
