Seeing Together: Multi-Robot Cooperative Egocentric Spatial Reasoning with Multimodal Large Language Models
The paper introduces CoopSR, an MLLM framework for multi-robot cooperative dynamic spatial reasoning using synchronized egocentric videos from moving robot teams. It integrates dynamics-aware frame sampling, spectral and physics-guided view fusion, and physics-aligned prompt distillation to enable fine-grained reasoning about spatial relationships, visibility, and coordination across different team sizes.