V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning

1 min read

V-JEPA 2 is a self-supervised joint-embedding predictive model pre-trained on over 1 million hours of internet video plus images. It scores 77.3 top-1 on Something-Something v2 motion understanding and 39.7 recall-at-5 on Epic-Kitchens-100 action anticipation, and does well on video question answering once aligned with an LLM. An action-conditioned version, V-JEPA 2-AC, post-trained on under 62 hours of unlabelled Droid robot video, lets Franka arms in two labs pick and place by planning towards image goals, with no data from those robots and no task-specific reward.

Relevance

JEPA-style latents show up across 2026 robot memory and world-model papers, including the surprise gate in Worth Remembering. Background for long-term autonomy and foundation models for robotics.