Papers
arxiv:2611.22275

Heave-Ho: Decentralized Long-Horizon Multi-Humanoid Collaboration

Published on Nov 17
Ā· Submitted by
Ting-Hsuan Chen
on Nov 18
Authors:
,
,
,
,
,
,
,
,
,

Abstract

A decentralized framework enables teams of humanoids to coordinate whole-body loco-manipulation over long horizons using low-level multi-agent post-training and a high-level vision-language planner with speech-based leader-follower synchronization.

Collaboration promises to extend humanoid capabilities beyond what individual robots can achieve. However, realizing this potential requires coordinated whole-body interactions, coherent long-horizon planning, and precise synchronization across robots. We introduce Heave-Ho, a decentralized framework for long-horizon collaborative multi-humanoid loco-manipulation. At the low level, we develop a multi-agent post-training framework that adapts a humanoid foundation tracking model to physically coupled interactions between humanoids, enabling coordinated decentralized control from local observations alone. At the high level, we train a vision-language planner and design a speech-based leader-follower coordination mechanism to synchronize decentralized execution without network communication. Trained with two identical humanoids, our system scales to teams of up to ten robots and generalizes to heterogeneous teams without retraining. To benchmark these capabilities, we introduce HH-Bench, comprising 12 tasks involving teams of 2 to 10 humanoids, including heterogeneous teams. Systematic evaluations show that Heave-Ho consistently outperforms baseline methods, while real-world experiments demonstrate robust decentralized collaboration across challenging tasks.

Community

Paper submitter

šŸ¤–šŸ¤šŸ¤– Heave-Ho: teams of humanoids collaborating on long-horizon tasks — fully decentralized, with no network communication between robots.

• Low level: multi-agent post-training adds a shared residual policy on top of a frozen humanoid foundation tracking model (SONIC), so each robot handles physically coupled whole-body interaction from its own local observations. 94% success in sim vs. ≤29% for baseline and ablated controllers.
• High level: a trained vision-language planner picks the next team skill, and the leader robot speaks it; followers recognize the command and execute in sync. 43/50 long-horizon trials vs. 21 for zero-shot GPT-6 (Astra), with ~120Ɨ lower per-decision latency.
• Trained with just 2 identical humanoids → scales to teams of up to 10, works with a different humanoid platform (G1 + PM01), and even with people as followers or leader.
• HH-Bench: 12 collaborative tasks for 2–10 humanoids, including table carry, narrow doors, L-shaped corridors, box handover, stair carry, and precise placement.

Videos: https://heaveho-humanoid.github.io/

Sign up or log in to comment

Models citing this paper 0

No model linking this paper

Cite arxiv.org/abs/2611.22275 in a model README.md to link it from this page.

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2611.22275 in a dataset README.md to link it from this page.

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2611.22275 in a Space README.md to link it from this page.

Collections including this paper 0

No Collection including this paper

Add this paper to a collection to link it from this page.