VITA: Zero-Shot Value Functions via Test-Time Adaptation of Vision-Language Models

Published in ICLR 2026, 2025