SCALE: Self-uncertainty Conditioned Adaptive Looking and Execution for Vision-Language-Action Models
cs.RO, cs.AI, cs.LG
Submitted: 2026-02-04
Updated: 2026-08-28
Comments: ICML 2026 Spotlight. Project page: https://dcahn12.github.io/projects/scale/
Code: https://github.com/snumprlab/scale
License: http://creativecommons.org/licenses/by/4.0/
The gist: Vision-Language-Action (VLA) models have emerged as a promising paradigm for general-purpose robotic control, with test-time scaling (TTS) gaining attention to enhance robustness beyond training.
Terminology
Abstract
Vision-Language-Action (VLA) models have emerged as a promising paradigm for general-purpose robotic control, with test-time scaling (TTS) gaining attention to enhance robustness beyond training. However, existing TTS methods for VLAs require additional training, verifiers, and multiple forward passes, making them impractical for deployment. Moreover, they intervene only at action decoding while keeping visual representations fixed-insufficient under perceptual ambiguity, where reconsidering how to perceive is as important as deciding what to do. To address these limitations, we propose SCALE, a simple inference strategy that jointly modulates visual perception and action based on 'self-uncertainty', inspired by uncertainty-driven exploration in Active Inference theory-requiring no additional training, no verifier, and only a single forward pass. SCALE broadens exploration in both perception and action under high uncertainty, while focusing on exploitation when confident-enabling adaptive execution across varying conditions. Experiments on simulated and real-world benchmarks demonstrate that SCALE improves state-of-the-art VLAs and outperforms existing TTS methods while maintaining single-pass efficiency.
Sources
- Deep Think with Confidence
- Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling
- Verifier-free Test-Time Sampling for Vision-Language-Action Models
- ReconVLA: Reconstructive Vision-Language-Action Model as Effective Robot Perceiver
- Evaluating Uncertainty and Quality of Visual Language Action-enabled Robots
- Policy Contrastive Decoding for Robotic Foundation Models
- AVA-VLA: Improving Vision-Language-Action models with Active Visual Attention
- Steering Vision-Language-Action Models as Anti-Exploration: A Test-Time Scaling Approach
- PEEK: Guiding and Minimal Image Representations for Zero-Shot Generalization of Robot Manipulation Policies
- EDT: Improving Large Language Models' Generation by Entropy-based Dynamic Temperature Sampling
- LIBERO-PRO: Towards Robust and Fair Evaluation of Vision-Language-Action Models Beyond Memorization
- InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models
- Confidence Calibration in Vision-Language-Action Models
Related papers
- FMT x: An Efficient and Asymptotically Optimal Extension of the Fast Marching Tree for Dynamic Replanning
- MPCFormer: A physics-informed data-driven approach for explainable socially-aware autonomous driving
- RoboLab: A High-Fidelity Simulation Benchmark for Analysis of Task Generalist Policies
- ARC: A Reasoning Recipe for Robot Foundation Models
- HRDexDB: A 4D Dexterous Grasping Dataset Across Human and Multiple Robot Embodiments
- APT: Action Expert Pretraining Improves Instruction Generalization of Vision-Language-Action Policies