ORCH: Organizational Principles Enable Collective Intelligence in Embodied AI
The paper says multi-agent robots performed better when their hierarchy was built around the task’s coordination pattern.
ORCH assigns roles and coordination layers by separating work that can run in parallel from work that has prerequisites. In 25 simulated wildfire-response missions with up to 50 heterogeneous agents, it beat four prior embodied multi-agent approaches on outcome, efficiency, exploration, and compute use. Human-designed ORCH structures raised final scores by 63.97% on average; LLM-generated ones raised them by 43.63%. The authors also report that larger models did not automatically produce better collective performance. ArXiv · AI/CL/LG's note
ORCH assigns roles and coordination layers by separating work that can run in parallel from work that has prerequisites. In 25 simulated wildfire-response missions with up to 50 heterogeneous agents, it beat four prior embodied multi-agent approaches on outcome, efficiency, exploration, and compute use. Human-designed ORCH structures raised final scores by 63.97% on average; LLM-generated ones raised them by 43.63%. The authors also report that larger models did not automatically produce better collective performance. ArXiv · AI/CL/LG's note
score 5