Unified framework for wide-range impedance control with high-fidelity motion mimicry on real hardware

Develop a unified control framework for humanoid robots that simultaneously enables wide-range impedance control (i.e., controllable interaction stiffness over a broad range) and preserves high-fidelity motion mimicry when deployed on real humanoid hardware. The framework should reconcile compliant whole-body interaction objectives with accurate tracking of reference motions during physical contact and environmental disturbances.

Background

Prior reinforcement learning approaches either emphasize motion tracking (often yielding stiff and potentially unsafe behavior) or focus on task-space force/impedance regulation without high-fidelity imitation of complex whole-body motions. Analytical methods provide impedance or admittance control formulations but are challenging to extend to floating-base humanoids while preserving motion style and tracking fidelity.

The paper motivates the need to combine the benefits of impedance control (safety, compliance, controllable interaction forces) with those of motion mimicry (faithful reproduction of humanlike reference motions) in a single deployable system. Although the authors propose SoftMimic as a step toward this goal, they explicitly identify the broader unification—covering a wide stiffness range while maintaining high-fidelity whole-body imitation on hardware—as an open challenge.