Get Started
Research questionHow can humanoid tracking policies learn pose transitions from unordered data without ordered motion clips?Humanoid tracking policies must explore physically plausible transitions between poses while following target motions. Unordered pose data describes pose variation but does not provide the temporal structure needed to guide those transitions directly.
Machine Learning
Reinforcement Learning
Robotics
Latest papersRecent research connected to this question, newest first.PFM-HR: Pose Flow Matching for Humanoid RobotsThe source studies PFM-HR, a reusable pose prior trained on large-scale unordered pose data and kept frozen across tracking tasks. Its Pose Geometry Score modulates the tracking reward, with evidence from single-motion and general motion-tracking experiments.research paper · Sep 3, 2026
Related questions
How can human motion be retargeted across different humanoid morphologies without hand-crafted mappings or losing pose detail?How can language-driven humanoid control follow instructions while staying physically plausible and stable over long horizons?How can direct vision-language-action robot policies capture multi-timescale dynamics without learning undesirable behavior from mixed-quality deployment trajectories?How can humanoid morphology and whole-body control be co-designed to reproduce human movement?
Home
Topics
Search
Library