Seeing Together: Multi-Robot Cooperative Egocentric Spatial Reasoning with Multimodal Large Language Models

Researchers introduced CoopSR, a new benchmark for multi-robot egocentric spatial reasoning, along with EgoTeam, a large QA dataset built from synchronized robot observations.

They also proposed SP-CoR, a model that fuses multi-view robot inputs with dynamics, spectral, and physics guidance to improve cooperative reasoning.

SP-CoR outperformed 22 baselines on spatial, temporal, visibility, and coordination tasks, and generalized to unseen team sizes.

The method also showed promising real-world performance with quadruped robots, extending egocentric AI from single-agent understanding to multi-robot collaboration.