Multi-Robot Collaborative Perception With Graph Neural Networks

Multi-robot systems such as swarms of aerial robots are naturally suited to offer additional flexibility, resilience, and robustness in several tasks compared a single robot by enabling cooperation among the agents. To enhance autonomous decision-making process situational awareness, multi-robot have coordinate their perception capabilities collect, share, fuse environment information agents an efficient meaningful way accurately obtain context-appropriate or gain resilience sensor noise failures. In this paper, we propose general-purpose Graph Neural Network (GNN) with main goal increase, tasks, robots' inference accuracy well failures disturbances. We show that proposed framework can address multi-view visual problems monocular depth estimation semantic segmentation. Several experiments both using photo-realistic real data gathered from multiple viewpoints effectiveness approach challenging conditions including images corrupted heavy camera occlusions