LeWAM: end-to-end JEPA lifts contact-rich task success from 28.6% to 89.7%, plans 32x faster
LeWAM · hf · 2026-10-14
Researchers from UC San Diego, ETH Zurich, UBC, and Brown introduce LeWAM, an end-to-end JEPA for control that jointly trains next-latent-state prediction with action flow matching under an isotropic Gaussian latent constraint, letting one model both execute policies directly and plan by scoring sampled actions. In visual planning it lifts contact-rich task success from 28.6% to 89.7% and long-horizon from 10.9% to 46.2%, while planning 32.3x faster; it also matches established methods on behavior cloning in simulation and on physical robots. Code and website are available.
More from Embodied
- Physical Intelligence researcher: robotics is in a 'Cambrian Explosion' — svlevine · 2026-10-14
- Robot demo nails 20/20 picks: nuts, gears and connectors in one run — KyleMorgenstein · 2026-10-14
- General Instinct partners with Physical Intelligence to deploy frontier models on real robots — svlevine · 2026-10-14
- Physical Intelligence opens Pi robotics model API beta, already in 150 factories — svlevine · 2026-10-14
- The 'sophisticated device' in spy case is a $14.90 Seeed Studio ESP32 hobby kit — wavefnx · 2026-10-14
- KILVO: robust multimodal odometry for humanoid robots fusing encoders, IMU, LiDAR and vision — rsasaki0109 · 2026-10-14