Motion Retargeting

Motion retargeting maps movement from one character or person to another body or robot with different proportions, joints, and abilities. Methods range from inverse kinematics and constrained trajectory optimization to learned graph, transformer, diffusion, and flow models; video reconstruction and physics simulation can help recover and refine motions. Work is evaluated on motion-capture and character-animation datasets, hand-object interaction tasks, and simulated or real robots, often testing whether transferred motions remain useful for control. A central challenge is balancing fidelity to the source with feasible movement and preserved contacts, especially when embodiments differ greatly or the source motion is incomplete.