Do Vision-Language-Action Models Mean What They Say? On the Role of Faithfulness in Embodied Reasoning
For embodied AI researchers, this work highlights the overlooked issue of faithfulness in reasoning traces and provides a method to improve it, though the gains are incremental over existing alignment techniques.
The paper identifies a gap between functional and faithful reasoning in Vision-Language-Action models, showing that state-of-the-art alignment strategies do not ensure faithfulness. The authors propose a learned critic (Pinocchio) to score faithfulness and use it as a reward signal in post-training, achieving a 4% and 18% improvement in faithfulness over baselines and 1.6x better responsiveness to out-of-distribution scenarios.
Embodied Chain-of-Thought has emerged as a promising mechanism to enhance robot decision-making and interpretability in black-box Vision-Language Action (VLA) models. However, whether this verbalized Chain-of-Thought truthfully reflects the policy's underlying decision process remains poorly understood. We distinguish between functional reasoning, in which reasoning improves task performance, and faithful reasoning, in which reasoning truly reflects the policy's internal decision process. We argue that SoTA alignment strategies offer a necessary but insufficient notion of faithfulness, admitting reasoning whose intermediate steps can mask the causal links in action prediction through confounding factors (e.g., reasoning that is ungrounded in the environment and internally disconnected or inconsistent), restricting policy generalization. We study this gap through a human evaluation of a SoTA reasoning model for autonomous driving, revealing an inconsistent coupling between reasoning quality and downstream trajectory improvement. We then operationalize a behavioral surrogate for embodied faithfulness through a learned critic, Pinocchio, scoring observation grounding and stepwise coherence, and use this critic as a dense reward signal in post-training an embodied policy with reinforcement learning. Across withheld driving benchmarks, our post-trained planner improves faithfulness by 4% and 18% over SoTA alignment and trajectory error post-training baselines, respectively, while maintaining competitive downstream task performance. Finally, on a synthetic out-of-distribution test set, post-training for faithfulness improves policy responsiveness to rare counterfactual scenarios by 1.6x that of a SoTA policy, suggesting that faithful reasoning traces contribute to more robust, generalizable, and interpretable embodied intelligence. Project page: https://mjf-su.github.io/pinocchio/