AI & ML New Capability

Optimizes diffusion models via Direct Preference Optimization (DPO) to generate human motion that is inherently executable by real humanoid robots.

arXiv · March 16, 2026 · 2603.13228

Yangsong Zhang, Anujith Muraleedharan, Rikhat Akizhanov, Abdul Ahad Butt, Gül Varol, Pascal Fua, Fabio Pizzati, Ivan Laptev

Why it matters

It closes the gap between 'pretty' generated animations and physically valid robotics, enabling zero-shot motion transfer to hardware like the G1 humanoid without post-hoc correction distortion.

From the abstract

Recent progress in text-conditioned human motion generation has been largely driven by diffusion models trained on large-scale human motion data. Building on this progress, recent methods attempt to transfer such models for character animation and real robot control by applying a Whole-Body Controller (WBC) that converts diffusion-generated motions into executable trajectories. While WBC trajectories become compliant with physics, they may expose substantial deviations from original motion. To a