TeamHOI: Learning a Unified Policy for Cooperative Human-Object Interactions with Any Team Size

TL;DR

TeamHOI introduces a scalable, unified policy framework enabling cooperative human-object interactions with any team size, leveraging Transformer-based coordination and motion priors for realistic, diverse behaviors.

Contribution

The paper presents a novel decentralized policy architecture using Transformers and a masked Adversarial Motion Prior to handle variable team sizes and promote realistic cooperative behaviors.

Findings

Achieves high success rates in cooperative carrying tasks with 2-8 agents.

Demonstrates coherent cooperation across diverse team sizes and object geometries.

Introduces a formation reward for stable team carrying behaviors.

Abstract

Physics-based humanoid control has achieved remarkable progress in enabling realistic and high-performing single-agent behaviors, yet extending these capabilities to cooperative human-object interaction (HOI) remains challenging. We present TeamHOI, a framework that enables a single decentralized policy to handle cooperative HOIs across any number of cooperating agents. Each agent operates using local observations while attending to other teammates through a Transformer-based policy network with teammate tokens, allowing scalable coordination across variable team sizes. To enforce motion realism while addressing the scarcity of cooperative HOI data, we further introduce a masked Adversarial Motion Prior (AMP) strategy that uses single-human reference motions while masking object-interacting body parts during training. The masked regions are then guided through task rewards to produce diverse and physically plausible cooperative behaviors. We evaluate TeamHOI on a challenging cooperative carrying task involving two to eight humanoid agents and varied object geometries. Finally, to promote stable carrying, we design a team-size- and shape-agnostic formation reward. TeamHOI achieves high success rates and demonstrates coherent cooperation across diverse configurations with a single policy.