Research Analyzer
← Back ICRA 2026

Spatio-Temporal Motion Retargeting for Quadruped Robots

Taerim Yoon, Dongho Kang, Seungmin Kim, Jin Cheng, Min Sung Ahn, Stelian Coros, Sungjoon Choi

PDF

AI summary

Key figure (auto-extracted from paper)
A two-stage spatio-temporal retargeting method successfully converts noisy, baseless keypoint trajectories into physically feasible motions, enabling agile quadruped robots to replicate diverse source movements in the real world.
Motion retargeting Quadruped robots Imitation learning Kinodynamic feasibility Reinforcement learning Spatio-temporal optimization

Problem

Existing motion retargeting methods often produce kinodynamically infeasible motions and struggle to adapt source data lacking global pose information, such as handheld videos, to robots with different morphologies.

Approach

The authors decouple retargeting into spatial and temporal stages: first reconstructing kinematically feasible whole-body poses from local keypoint trajectories, then optimizing motion timing via a nested model-based controller to satisfy dynamic constraints.

Key results

  • Generates kinodynamically feasible whole-body motions from baseless keypoint trajectories
  • Eliminates foot sliding and preserves contact schedules during retargeting
  • Enables successful real-world deployment of agile skills across four quadruped robots with varying dimensions
  • Demonstrates terrain-aware retargeting for complex maneuvers like backflipping on elevated surfaces

Why it matters

Bridges the gap between diverse motion capture sources and legged robot deployment, accelerating the development of agile, morphology-adaptive robotic behaviors.

Abstract

This work presents a motion retargeting approach for legged robots, aimed at transferring the dynamic and agile movements to robots from source motions. In particular, we guide the imitation learning procedures by transferring motions from source to target, effectively bridging the morphological disparities while ensuring the physical feasibility of the target system. In the first stage, we focus on motion retargeting at the kinematic level by generating kinematically feasible whole- body motions from keypoint trajectories. Following this, we refine the motion at the dynamic level by adjusting it in the temporal domain while adhering to physical constraints. This process facilitates policy training via reinforcement learning, enabling precise and robust motion tracking. We demonstrate that our approach successfully transforms noisy motion sources, such as hand-held camera videos, into robot-specific motions that align with the morphology and physical properties of the target robots. Moreover, we demonstrate terrain-aware motion retargeting to perform BackFlip on top of a box. We successfully deployed these skills to four robots with different dimensions and physical properties in the real world through hardware experiments.

Index terms

Legged Robots Learning from Demonstration Motion Control Motion Retargeting

Related papers