A new white paper outlines a framework that decouples motion planning from tracking to preserve professional athletic style in humanoid robots. The approach trains trackers on randomized speeds and pairs them with a motion‑speed adapter, reducing sim‑to‑real errors.
Core Facts: Decoupled Planning & Speed‑Aware Tracking
- High‑level planner and low‑level tracker are separated to balance style and task performance.
- Reference motions are reconstructed from broadcast video and motion‑capture data, then retargeted for safe robot execution.
- Trackers are trained on randomized execution speeds, while a learned motion‑speed adapter in the planner adapts plans accordingly.
- Validation on two humanoid platforms, including full‑size in‑the‑wild play without any motion‑capture system.
Background: Bridging the Sim‑to‑Real Gap in Ball‑Sports Robotics
Combining a professional athlete’s fluid motion with reliable task execution has long been a hurdle for humanoid ball‑sports robots. Traditional approaches fuse planning and tracking, leading to compounded errors when moving from simulation to physical hardware. This white paper addresses the trade‑off by decoupling the two stages and introducing adaptive speed awareness, a strategy that aligns simulated motion with real‑world dynamics.
Future Impact: Toward Real‑World Pro‑Style Play
The adaptive framework promises to accelerate the deployment of humanoid robots in competitive sports and entertainment. By closing the sim‑to‑real gap, developers can now prototype complex athletic routines in simulation and transfer them to physical robots with minimal tuning, opening avenues for advanced training tools, live demonstrations, and even commercial sports robotics.