Test-Time Adaptation for Generalizable Task Progress Estimation

Christos Ziakas, Alessandra Russo

Keywords: Test-Time Adaptation, Task Progress Estimation, Vision-Language Models, Meta-Learning

TL;DR: A test-time adaptation method for task progress estimation that generalizes to unseen environments.

Abstract: We propose a test-time adaptation method that enables a progress estimation model to adapt online to the visual and temporal context of test trajectories by optimizing a learned self-supervised objective. To this end, we introduce a gradient-based meta-learning strategy that explicitly trains the progress estimator to learn to adapt at test time, inducing reliance on semantic content rather than temporal order. Our test-time adaptation method generalizes from a single training environment to diverse out-of-distribution tasks, environments, and embodiments, outperforming the state-of-the-art in-context learning approach using autoregressive vision-language models.

Submission Number: 23