Comparison of Channel Attention Mechanisms and Graph Attention Mechanisms Applied in Multi-Robot Path Planning Based on Graph Neural Networks

Abstract

Graph Neural Networks (GNNs) are applied to multi-robot path planning due to their advantages in learning communication strategies in decentralized multi-intelligence systems, and to prioritize the important information, we introduce the channel attention mechanism and the graph attention mechanism in GNNs. The channel attention mechanism automatically obtains the importance of each feature channel through network learning, assigns different weight coefficients to each channel, and calculates the importance of each channel of the input image so as to enhance the important features and suppress the unimportant features. Unlike the channel attention mechanism, which calculates the importance of each channel to improve the feature representation, the graph attention mechanism is a new type of neural network that operates on graph-structured data and uses a masked self-attention layer to address the shortcomings of graph convolution-based methods.