Search papers, labs, and topics across Lattice.
This paper addresses the fragility of generalization in Vision-Language-Action (VLA) models by introducing the SVA framework, which enhances action evaluation without compromising the model's generalist capabilities. Through a diagnostic pass@k study, the authors reveal that frozen VLAs exhibit competent behaviors, with success rates improving significantly from 33% to 92% as the evaluation depth increases. By employing Monte-Carlo tree search for action evaluation and distilling this knowledge into a lightweight Q-value model, SVA allows for improved task success rates and generalization on unseen tasks, even enabling a smaller VLA model to outperform a larger counterpart.
A lightweight Q-value model can boost a 9B VLA's performance beyond that of a 27B model while reducing inference latency by 27%.
Vision-Language-Action (VLA) models acquire broad embodied capabilities through large-scale pretraining, yet their generalization remains far more fragile than that of LLMs and VLMs. The prevailing remedy, post-training via supervised fine-tuning or reinforcement learning, improves task-specific performance but narrows the generalist capability that makes pretraining valuable. We identify a key bottleneck: VLA failures stem not only from action generation but also from action evaluation. A diagnostic pass@k study confirms that frozen VLAs already contain competent behaviors in their output distribution, with overall success rates rising from 33% at pass@1 to 92% at pass@32. Inspired by this, we propose SVA (Search, Value, and Act), a simple framework that equips frozen VLA policies with long-term consequence awareness. SVA first uses Monte-Carlo tree search in simulation to fully explore the VLA's output distribution and collect diverse trajectories annotated with empirical returns; this knowledge is then distilled into a lightweight Q-value model that predicts the expected consequence of candidate actions; at deployment, the frozen VLA proposes multiple candidates and the evaluator selects the one with the highest uncertainty-regularized Q-value, requiring no simulator access. By decoupling action proposal from consequence evaluation, SVA preserves the generalization capacity of the VLA backbone while substantially improving task success rates. Experiments across embodied benchmarks show that SVA consistently improves generalization on unseen tasks and exhibits strong test-time scaling behavior. Strikingly, SVA enables a 9B VLA to outperform a 27B VLA by 7 points at 27% lower inference latency, suggesting that scaling test-time evaluation is more cost-effective than scaling model size.
