CoSaR framework enables multi-robot teams to coordinate city-scale navigation safely

A team of researchers has formalized a new problem setting called Cooperative Spatial Intelligence, where multiple decentralized embodied agents must coordinate in large, city-scale outdoor environments. They introduced the Sentinel Challenge benchmark, which requires agents to communicate via natural language to agree on a mutually safe meeting point while navigating around dynamic sentinel patrols. The agents are equipped only with coarse spatial information, forcing them to rely on reasoning and communication.

To solve this, the team developed CoSaR (Cooperative Spatial Reasoning and Planning), a framework that combines the high-level planning and language capabilities of foundation models with the precision of classical spatial navigation algorithms. CoSaR enables agents to exchange situational updates, reason over evolving spatial constraints, and collaboratively replan trajectories. Evaluated across 14 city-level scenes with 3-5 agents, CoSaR consistently outpaced existing methods, achieving faster gathering times, shorter overall path lengths, and improved safety. The researchers argue that integrating dynamic communication with spatial reasoning is essential for robust multi-agent cooperation. Code and the Sentinel Challenge are publicly available.