A spatio-temporal graph reinforcement learning-based multi-robot trajectory planning method for small-scale fading in underground communications | Synapse