DM$^3$-Nav: Decentralized Multi-Agent Multimodal Multi-Object Semantic Navigation

Researchers at Northeastern University have unveiled DM³-Nav, a fully decentralized multi-agent system that enables teams of robots to navigate complex environments and locate multiple objects using only local, ad-hoc peer-to-peer communication. Unlike traditional multi-robot systems that rely on a central coordinator or shared global map, DM³-Nav operates without any global state or synchronization. Each robot runs autonomously, exchanging local maps, goal status, and navigation intent with nearby teammates as needed. An implicit task allocation mechanism—combining intent broadcasting with distance-weighted frontier selection—reduces redundant exploration while preserving fully decentralized operation.

The system was rigorously evaluated on HM3DSem scenes using the HM3Dv0.2 and GOAT-Bench datasets, demonstrating that DM³-Nav matches or exceeds the performance of centralized and shared-map baselines. Critically, it eliminates the single points of failure inherent in centralized architectures. The team also validated DM³-Nav in a real-world office environment using two mobile robots, relying entirely on onboard sensing and computation. This approach could dramatically improve the robustness and scalability of robot swarms in search-and-rescue, warehouse logistics, and planetary exploration, where communication infrastructure is limited or unreliable.