H2GNN: Hierarchical-Hops Graph Neural Networks for Multi-Robot Exploration in Unknown Environments

Abstract

Multi-robot coarse-to-fine exploration in unknown environments makes great sense in many application fields like search and rescue. For different stages of the task, robots need to extract information from the environment discriminately, which can improve their decision-making capability. To this end, we present the Hierarchical-Hops Graph Neural Networks (H2GNN) to enable robots to targetedly integrate the key information of the graph-represented environment, which distinguishes the importance of information from different hops around robots based on the multi-head attention mechanism. And in order to improve the efficiency of cooperation, we utilize multi-agent reinforcement learning (MARL) to help robots to learn collaborative strategies implicitly. We conduct experiments to verify our proposed method in a simulation environment, and the experimental results demonstrate that the H2GNN significantly improves the multi-robot exploration performance in unknown environments.