Humanoid robots hold promise as general-purpose agents in human-centered environments, but adapting generalist vision-language-action (VLA) policies to real-world humanoid loco-manipulation remains a significant challenge, according to a new arXiv paper.

The paper, titled HAF-VLA, proposes a Hierarchical Action Flow Framework to tackle this problem. The abstract indicates the framework is designed to bridge the gap between generalist VLA policies and the specific demands of humanoid locomotion and manipulation in real-world settings.

The abstract does not provide implementation details, but the title suggests a hierarchical structure for action generation. The paper is available as arXiv:2610.26584v1.