arXiv preprint reports organizational hierarchy boosts embodied multi-agent teams

A preprint posted to arXiv describes ORCH, a framework that builds task-specific hierarchical organizations for large collectives of embodied artificial agents by combining pooled interdependence for concurrent work with sequential interdependence for tasks governed by prerequisites. The authors evaluated teams of up to 50 heterogeneous agents using eight large language models across 25 wildfire-response missions covering reconnaissance, rescue, transportation, resource management, containment and suppression. They report that human-designed ORCH organizations improved final score by 63.97% and execution efficiency by 74.29% on average relative to four prior embodied multi-agent frameworks, while language-model-generated organizations improved those measures by 43.63% and 52.53%. The authors state the advantages persisted across missions and underlying models, and that collective performance was not monotonically determined by model scale.