Temporal Difference Calibration in Sequential Tasks: Application to Vision-Language-Action Models

TL;DR

This paper introduces a novel temporal difference calibration method for sequential tasks in vision-language-action models, linking uncertainty calibration with reinforcement learning to improve robot decision-making.

Contribution

It develops a sequential extension of the Brier score and demonstrates that TD value estimation can serve as an effective calibration mechanism over time.

Findings

TD calibration improves model performance on robot data

Calibrated action probabilities provide competitive uncertainty estimates

The approach bridges uncertainty calibration with reinforcement learning

Abstract

Recent advances in vision-language-action (VLA) models for robotics have highlighted the importance of reliable uncertainty quantification in sequential tasks. However, assessing and improving calibration in such settings remains mostly unexplored, especially when only partial trajectories are observed. In this work, we formulate sequential calibration for episodic tasks, where task-success confidence is produced along an episode, while success is determined at the end of it. We introduce a sequential extension of the Brier score and show that, for binary outcomes, its risk minimizer coincides with the VLA policy's value function. This connection bridges uncertainty calibration and reinforcement learning, enabling the use of temporal-difference (TD) value estimation as a principled calibration mechanism over time. We empirically show that TD calibration improves performance relative to the state-of-the-art on simulated and real-robot data. Interestingly, we show that when calibrated using TD, the VLA's single-step action probabilities can yield competitive uncertainty estimates, in contrast to recent findings that employed different calibration techniques.