V-JEPA 2-AC does zero-shot robot pick-and-place after only 62 hours of robot video
TheTuringPost · x · 2026-11-21
From TheTuringPost's thread: V-JEPA 2 (June 2025) was pretrained on 1M+ hours of internet video. Its action-conditioned variant, V-JEPA 2-AC, was post-trained on under 62 hours of robot video and performs zero-shot pick-and-place on Franka arms in two labs, planning toward image goals. Together with DINO-WM and PLDM, it established the recipe of pretrained encoder + predictor + planner—JEPA world models now act.
More from Embodied
- NVIDIA ships Alpamayo 2 Edge, a 2B-parameter in-vehicle reasoning VLA model — NVIDIAAI · 2026-11-21
- LIFE profile: Noland Arbaugh, Neuralink's first brain implant patient — kyliebytes · 2026-11-21
- Robotics supply chains can't be reshored — they still need to be built from scratch — theo_gervet · 2026-11-21
- Rysen's Apex Hand does tweezer-level fine manipulation via MANUS glove teleop — chris_j_paxton · 2026-11-21
- Reverse-engineering Apple Watch: a single Claude Code prompt builds a physics watch face in 30 minutes — mtizard · 2026-11-21
- Robotics Founder: Housekeeping Bots Should Target Mundane Upkeep, Not Party Tricks — Ben_Reinhardt · 2026-11-21