Kinodynamic Motion Retargeting for Humanoid Locomotion via Multi-Contact Whole-Body Trajectory Optimization

TL;DR

Experimental results demonstrate that KDMR significantly outperforms purely kinematic methods, yielding dynamically viable reference trajectories that accelerate policy convergence and enhance overall locomotion stability.

Abstract

We present the KinoDynamic Motion Retargeting (KDMR) framework, a novel approach for humanoid locomotion that models the retargeting process as a multi-contact, whole-body trajectory optimization problem. Conventional kinematics-based retargeting methods rely solely on spatial motion capture (MoCap) data, inevitably introducing physically inconsistent artifacts, such as foot sliding and ground penetration, that severely degrade the performance of downstream imitation learning policies. To bridge this gap, KDMR extends beyond pure kinematics by explicitly enforcing rigid-body dynamics and contact complementarity constraints. Further, by integrating ground reaction force (GRF) measurements alongside MoCap data, our method automatically detects heel-toe contact events to accurately replicate complex human-like contact patterns. We evaluate KDMR against the state-of-the-art baseline, GMR, across three key dimensions: 1) the dynamic feasibility and smoothness of the retargeted motions, 2) the accuracy of GRF tracking compared to raw source data, and 3) the training efficiency and final performance of downstream control policies trained via the BeyondMimic framework. Experimental results demonstrate that KDMR significantly outperforms purely kinematic methods, yielding dynamically viable reference trajectories that accelerate policy convergence and enhance overall locomotion stability. Our end-to-end pipeline will be open-sourced upon publication.

Figures (7)

• Figure 1: The presented Kinodynamic Motion Retargeting (KDMR) framework translates human motion references and ground reaction forces into retargeted motions that are dynamically feasible and respect multi-contact holonomic constraints. We further demonstrate that policies trained using our retargeted motions exhibit superior sample efficiency than those trained on kinematic data alone.
• Figure 2: Natural human walking, which we call multi-contact locomotion, is characterized by a heel-to-toe rolling motion. The diagram illustrates one complete gait cycle, beginning and ending with right heel strike.
• Figure 3: Algorithm overview: The KDMR algorithm converts time-series marker and ground reaction force data into joint position targets for the humanoid robot. This is accomplished in two steps: (1) Data Processing, which converts spatial marker trajectories and GRF data to human model poses and contact states; (2) Kinodynamic Motion Retargeting, which converts a sequence of human poses and contacts to a corresponding robot trajectory.
• Figure 4: Illustration of the human GRF pattern used to estimate the contact sequence for steady-state walking.
• Figure 5: Comparison of retargeted trajectories between KDMR and GMR over a selected time segment (32–40 s) from the full motion sequence. (a) Robot base position (x, y, z) and orientation (roll, pitch, yaw). (b) Corresponding base linear and angular velocities. (c) Joint position of the left leg (hip, knee, and ankle joints). (d) Corresponding joint velocity. The $\theta_{Li}$ and $\dot{\theta}_{Li}$indicate the joint position and velocity of $i$-th joint.