KAIST's FastOPD cuts VLA robot inference latency by 78% via on-policy distillation
kaist-ai · hf · 2026-10-08
KAIST AI introduces FastOPD, a foundation-to-lightweight framework for on-policy distillation of Vision-Language-Action models. It adapts a flow map for single-state teacher supervision plus a self-consistency objective, with theory showing the student recovers an ideal few-step teacher distribution. On LIBERO it retains 84% of π₀.₅'s performance with two inference steps, cutting latency by 78.1% and beating prior few-step distillation baselines. With LingBot-VLA as teacher it lifts single-step success on RoboTwin 2.0 by 15.9 points, and a student distilled from MolmoAct2 was deployed on a real robot.
More from Embodied
- Undergrad attends first IROS, gets hands-on with new humanoid robots — heatherknight · 2026-10-08
- Desktop robots MOSS and Vita star in an 80-minute robotics seminar at Thammasat University — StewartalsopIII · 2026-10-08
- Nvidia bets billions on physical AI, expanding Halos safety system to robots — Ars Technica AI · 2026-10-08
- Singapore debuts real-life Rock 'Em Sock 'Em robot boxing match — DJiafei · 2026-10-08
- LIBERO-MAX: mid-task changes cut success rates for all 14 robot policies by 11-25.7 points — DJiafei · 2026-10-08
- Meta's 8B RoboJEPA shows robot world models have scaling laws, trained on 15k hours of video — ylecun · 2026-10-08