ADAPT at CoRL 2026: end-to-end diffusion policy for real-time humanoid language control
breadli428 · x · 2026-11-10
ADAPT, a humanoid language-control work from breadli428 and team, will be presented at CoRL 2026 (Nov 11, poster #174).
- Unlike dominant text-to-motion pipelines that generate kinematic motions for a separate tracker, ADAPT solves humanoid language control end-to-end in closed loop, continuously responding to changing commands
- Core technique: a robust and steerable diffusion-based motion prior for real-time humanoid control
- Interactive browser demo: a MuJoCo playground (29 DOF) running the diffusion policy live on-device, supporting text-driven control and goal-reaching mode (175MB first load)
More from Embodied
- Google launches Gemini Robotics 2.5, first foundation model accelerated by Gemini 4 Argon in recursive self-improvement — davidpattersonx · 2026-11-11
- Zipline CEO on starting in Rwanda, $21.5B valuation and 50B deliveries — zealcaiden · 2026-11-11
- Muse's Apple Watch integration hands-on: a preview of the upcoming Muse Charm — altryne · 2026-11-11
- Origami's DexUMI exoskeleton glove for dexterous robot data collection — chris_j_paxton · 2026-11-11
- Hedra launches Model Factory, an autonomous training platform opening frontier-style self-improvement loops to robotics and industry — EXM7777 · 2026-11-11
- Walden Robotics at CoRL: Russ Tedrake keynote on learning in factories — adnothing · 2026-11-11