Spatio-Temporal Motion Retargeting for Quadruped Robots for IEEE T-RO

Teaching a quadruped to dance, jump, and trot from a handheld video. This T-RO paper retargets motion from noisy keypoint trajectories to robot-specific whole-body movements, then refines timing to respect physical constraints. Six different motions on two different quadrupeds, all in the real world. The video demonstrates dynamic motion-retargeting pipeline for legged robots, transferring agile movements from diverse sources despite differences in morphology and physical limits. It employs a two-stage process—first mapping keypoint trajectories to kinematically feasible whole-body motions, then refining them through temporal adjustments to enforce physical realism. Noisy inputs (e.g., hand-held video captures) are cleaned and reshaped into robot-specific motions that respect each platform’s dynamics and structure. Terrain awareness lets the robot tackle demanding maneuvers—such as backflipping off an elevated surface—while preserving fidelity to the original motion. Finally, hardware tests on four different legged robots confirm the approach’s versatility and robustness in real-world settings. Presented at #ICRA2026 Paper URL: https://ieeexplore.ieee.org/document/... Paper Title: Spatio-Temporal Motion Retargeting for Quadruped Robots Authors: Taerim Yoon, Dongho Kang, Seungmin Kim, Jin Cheng, Min Sung Ahn, Stelian Coros & Sungjoon Choi #MotionRetargeting #LeggedRobots #ReinforcementLearning