Foundation trackers improve leg motion on uneven ground with geometry awareness
GRAFT: Geometry-Aware Residual Adaptation for Foundation Trackers
Robotics
Summary
Tracking whole-body motions for humanoid robots is easier on flat surfaces because existing models don’t adjust well to uneven terrain like stairs or gaps. The authors propose a method called GRAFT that keeps the main motion model fixed but adds a small, smart adjustment that changes foot movements based on terrain shape. This makes robots better at placing feet properly without retraining the entire system, saving time and effort. They tested this approach in simulations and on a real robot, showing it works with onboard sensors to handle rough ground.
What this means in practice
- •For robotics engineers: Make humanoid or legged robots walk smoothly over stairs, gaps, or uneven surfaces without retraining the entire motion model by adding terrain-specific foot adjustments.
- •For robot operators: Improve real-time control of robots in field environments by adapting existing motion planning to local terrain using onboard depth sensors and lightweight correction modules.
Authors
Liang Li, Guoyao Zhang, Xingyu Chen, Sikai Wu, Qianyang Wu, Yuhang Lin, Renjing Xu, Mingliang Zhou, Diyun Xiang
Abstract
Humanoid behavior foundation models (BFMs) provide reusable whole-body motion priors: given a kinematic reference, they generate coordinated joint commands for motion replay, teleoperation, and VLA control. However, trained primarily on flat-ground motion, BFMs lack an explicit mechanism to adapt foot placement, swing clearance, and contact timing to local terrain. Many terrain-aware BFM approaches address this limitation by retraining or fine-tuning the tracker together with perception, incurring substantial data and computation costs while coupling ordinary tracking to the perception pathway. To bridge this gap, we propose GRAFT, a geometry-aware residual adaptation framework for frozen behavior trackers. GRAFT preserves the BFM parameters and motion-reference interface, while a lightweight adapter adds terrain-conditioned corrections only to lower-body execution targets. The adapter is trained with task and foot-geometry rewards, regularized to suppress unnecessary corrections on flat ground, and distilled from privileged terrain observations to onboard depth. Because the frozen BFM already provides whole-body reference tracking and nominal locomotion, GRAFT reduces terrain learning to a focused execution-adaptation problem. We instantiate the framework on four frozen trackers and evaluate stairs, gaps, and platforms in simulation. We further validate it through real-world deployment on a Unitree G1, where the adapted SONIC tracker traverses terrain with onboard depth and unmodified references.