The domains of transport and logistics are increasingly relying on autonomous mobile robots for the handling and distribution of passengers or resources. At large system scales, finding decentralized path planning and coordination solutions is key to efficient system performance. We tackle scalability challenges in decentralized path planning for urban transit and logistics. We utilize Graph Neural Networks (GNNs), adept at processing topological models and non-Euclidean data, to build an information-sharing mechanism within a decentralized multi-robot system. Combining GNNs with imitation learning, we train a low-power framework approximating optimal algorithms without high computational demands. Our physical experiments involve a purely visual, machine learning-assisted navigation technology. This technology encodes relevant viewpoint information from panoramic cameras scattered in the environment through convolutional networks and communicates it to mobile robots via a GNN. Trained via imitation learning, the model effectively guides robots to their destinations, achieving performance close to the best motion algorithms. Our experiments demonstrate its universality in guiding robots in previously unseen environments with various sensor layouts.
Our work demonstrates the potential of integrating advanced computer vision and machine learning techniques to enhance the capabilities of embodied artificial intelligence in complex and dynamic environments, improving real-time decision-making, navigation, and overall system performance, and how multi-robot system will bring impact into industry and our daily life.