ULTRA unifies motion tracking and sparse goal control in a single policy for humanoid robots, combining physics-driven motion retargeting, policy distillation, and RL fine-tuning, validated on Unitree G1 with first-person point cloud perception.
The article surveys motion retargeting, describing its importance for virtual characters, outlining traditional geometric methods and recent deep‑learning approaches, presenting a customized Interaction Mesh‑based solution from Kuaishou Y‑tech, and discussing performance, limitations, and future research directions.
Motion retargeting, also known as motion adaptation, transfers source motion to arbitrary virtual characters while preserving semantic features and natural flow, and this article reviews traditional geometry‑based methods, recent deep‑learning approaches, and Kuaishou Y‑Tech’s optimized pipeline that balances quality, contact preservation, and real‑time performance.
