Graph neural network approach for decentralized multi-robot coordination

Today’s urban rail transit and warehousing logistics increasingly rely on automated mobile robots to handle and distribute passengers or resources. However, as the scale of the system expands dramatically, decentralized rmutli-obot path algorithms will be key to improving system effectiveness. In recent years, Graph Neural Networks have become increasingly active in the field of machine learning, used to process topological models or non-Euclidean data. This report will focus on what Graph Neural Networks are, how to build an information sharing mechanism within a decentralized multi-robot system through Graph Neural Networks, and combine imitation learning to train a decentralized, low-power framework to approximate the performance of optimal solution algorithms that have higher demands on computational power and time. In terms of physical experiments, we propose a purely visual machine learning-assisted navigation technology. This technology first encodes relevant viewpoint information of panoramic cameras scattered in the environment through a convolutional network, and then communicates it to mobile robots through a graph neural network. The model will obtain performance close to the best motion algorithm through imitation learning training, thus effectively guiding robots in the environment to their destination. Our experiments have proven its universality in guiding robots in previously unseen environments with various sensor layouts.

Post Talk Link:  Click Here 

Passcode: .2jo$BXZ