JEPA-VLA: Video Predictive Embedding is Needed for VLA Models

arXiv · 2026-02-12 · Low-Level Learning-Based Action Modelling / Input Modelling / 2D Vision Language Action Models with Latent Learning