OmniContact: Chaining Meta-Skills via Contact Flow for Generalizable Humanoid Loco-Manipulation

2026-06-24

AI Summary

This work addresses the challenges of robust execution, seamless chaining, and autonomous recovery of meta-skills in long-horizon loco-manipulation tasks for humanoid robots. To this end, the authors propose a contact flow–based hierarchical framework: at the low level, CF-Track learns a unified policy over a library of meta-skills, while at the high level, CF-Gen heuristically generates future contact flow sequences to enable flexible skill composition and self-recovery. By leveraging contact flow as a compact and interpretable representation of physical interaction, the approach bridges high-level task planning with low-level motion control and facilitates semantic integration with vision-language models. Experiments demonstrate success rates of 98.7% and 76.5% on the Carry Box and Push-Stack Boxes tasks, respectively, with meta-skill and skill-chain performance improvements of 40.9% and 66.5% over baseline methods.

Abstract

Learning long-horizon humanoid loco-manipulation poses a dual challenge: it requires not only the robust execution of meta-skills but also their seamless, closed-loop chaining equipped with autonomous recovery. Existing approaches remain limited: explicit humanoid-object interaction representations offer precision but are notoriously difficult for high-level planning, whereas implicit skill embeddings are compact but lack the interpretability required for reliable composition. We propose \ours, a hierarchical framework centered on \textbf{contact flow (CF)}, a compact representation consisting of key body trajectories and time-series binary contact signals. Leveraging this shared interface, our low-level policy \textbf{CF-Track} learns a unified library of loco-manipulation skills, while our high-level module \textbf{CF-Gen} heuristically synthesizes future contact-flow sequences. To support this setting, we additionally collect the OmniContact dataset, a MoCap-based HOI corpus for humanoid loco-manipulation (Appendix~\ref{sec:dataset}). Together, they enable robust execution, autonomous failure recovery, and flexible composition of meta-skills for long-horizon tasks. Experiments show that OmniContact achieves \(98.7\%\) success on \textit{Carry Box} and \(76.5\%\) on \textit{Push-Stack Boxes}, outperforming prior baselines by average margins of \(40.9\%\) in meta-skill and \(66.5\%\) in skill chaining. Besides, our framework naturally integrates with VLMs for semantic task decomposition, enabling complex, semantically grounded loco-manipulation behaviors, such as arranging scattered boxes into a heart shape.

Problem

• humanoid loco-manipulation
• meta-skills
• skill chaining
• contact representation
• long-horizon tasks

Innovation

• Contact Flow
• Loco-Manipulation
• Meta-Skill Chaining
• Humanoid Robotics
• Hierarchical Policy