Abstract
This paper addresses the task of joint multi-agent perception and planning,especially as it relates to the real-world challenge of collision-freenavigation for connected self-driving vehicles. For this task, severalcommunication-enabled vehicles must navigate through a busy intersection whileavoiding collisions with each other and with obstacles. To this end, this paperproposes a learnable costmap-based planning mechanism, given raw perceptualdata, that is (1) distributed, (2) uncertainty-aware, and (3)bandwidth-efficient. Our method produces a costmap and uncertainty-awareentropy map to sort and fuse candidate trajectories as evaluated acrossmultiple-agents. The proposed method demonstrates several favorable performancetrends on a suite of open-source overhead datasets as well as within a novelcommunication-critical simulator. It produces accurate semantic occupancyforecasts as an intermediate perception output, attaining a 72.5% averagepixel-wise classification accuracy. By selecting the top trajectory, themulti-agent method scales well with the number of agents, reducing the hardcollision rate by up to 57% with eight agents compared to the single-agentversion.