V2X User Scheduling for Collaborative Perception Based on Reinforcement Learning

Collaborative perception aims at addressing occlusion in stand-alone perception and improving the perception accuracy of connected automated vehicles (CAVs) by sharing perceptual information between different CAVs and roadside units (RSU) through vehicular networks. In this paper, we investigate the user scheduling problem in vehicular networks and develop a novel scheduling algorithm based on deep Q-network (DQN). Our algorithm directly optimizes the collaborative perception accuracy by simultaneously incorporating the spatial confidence and the channel state information. Simulation results demonstrate that our proposed algorithm achieves higher perception accuracy compared with round robin and the maximum rate scheduling methods under various settings.