Summary
HANDOFF is a humanoid whole-body controller that uses a compact, explicit command interface designed to be intuitive, general, modular, and expressive enough for diverse manipulation skills. The method distills three complementary specialist teachers—whole-body motion tracking, safe filtered data, and motion plus fall recovery—into a single mixture-of-experts student policy via multi-teacher KL distillation under a context-conditioned gating scheme. The command space choice is critical for real-world humanoid deployment, bridging task-level planning and whole-body control; unlike existing controllers that require dense kinematic or spatial references, HANDOFF's interface can be synthesized directly from task semantics. Evaluated on the Unitree G1 humanoid robot, HANDOFF matches state-of-the-art velocity tracking performance and provides one of the largest robust manipulation workspaces reported. The paper (arXiv:2506.08300) was published on June 11, 2025 by Lizhi Yang, Junheng Li, and Nehar Poddar, and addresses a key gap in humanoid robotics deployment.
Paper Overview
Research Area: Machine Learning
Authors: Lizhi Yang, Junheng Li, Nehar Poddar
Published: 2025-06-11
arXiv: 2506.08300
Abstract
For a humanoid robot to be deployed in the real world, the choice of command space is crucial—the command space serves as the interface between task-level planning and whole-body control. Existing whole-body controllers typically require dense kinematic or spatial references, which planners struggle to synthesize from task semantics alone.
The authors propose a compact, explicit interface that is intuitive, general, modular, and expressive enough to represent diverse manipulation skills. To realize this, they introduce HANDOFF—a single humanoid whole-body controller distilled from three complementary specialist teachers into a mixture-of-experts student:
- Whole-body motion tracking expert
- Safe filtered data expert
- Motion and fall recovery expert
Distillation is performed via multi-teacher KL distillation under a context-conditioned gating scheme.
Results
On the Unitree G1 humanoid robot, HANDOFF:
- Matches state-of-the-art velocity tracking performance
- Offers one of the largest robust manipulation workspaces
Links
- Paper: https://arxiv.org/abs/2506.08300
---
*Auto-collected on 2025-06-11*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/177980874