ADAPT: Diffusion action priors enable robust text-driven humanoid whole-body control
breadli428 · x · 2026-11-03
Researchers led by Yan Wu and Chenhao Li, with Marco Hutter and Siyu Tang among co-authors, present ADAPT, an end-to-end framework for interactive text-conditioned humanoid whole-body control, accepted at CoRL 2026 (Session 3, Poster #174).
Key points:
- Unlike dominant text-to-motion pipelines that generate kinematic motions for a separate tracker, ADAPT uses closed-loop control: the robot continuously responds to changing commands while maintaining balance, natural motion, and smooth transitions
- A diffusion-based action prior is learned from text-labeled humanoid state-action trajectories, letting diverse motion skills execute directly from language commands
- A lightweight residual RL policy is trained atop the frozen diffusion controller for long-horizon robustness and smooth prompt switching
- The same diffusion policy is reusable as a steerable text-conditioned motion prior for downstream task adaptation (e.g. "run" to goal vs "walk while bending over" to goal), preserving style
Paper: arXiv 2609.00677
Related event: ETH Zurich's ADAPT Enables Real-Time Language Control of Humanoid Robots(3 posts)→
More from Embodied
- ADAPT architecture breakdown: diffusion Transformer plus residual RL in four stages — breadli428 · 2026-11-03
- ADAPT: ETH Zurich's 50Hz end-to-end text-driven humanoid control with diffusion priors — breadli428 · 2026-11-03
- Xiaomi open-sources Robotics-1, a robot foundation model trained on 100K hours of trajectories — CyberRobooo · 2026-11-03
- Factory photos show desktop robot modules in volume assembly, project hits production stage — victormustar · 2026-11-03
- Machenta demos instant robot learning from passersby at London King's Cross — Ed__Johns · 2026-11-03
- PRIMA animal 3D mesh recovery accepted to WACV2027 with 9K pseudo-3D dataset — TrackingActions · 2026-11-03