Robust 4D awareness via diffusion adaptation over Connected and Autonomated vehicles

In this paper, we focus on enhancing the 4D situational awareness in a swarm of Connected and Automated vehicles, employing information diffusion against unsuccessful object detection. More specifically, in the perception layer we exploit model compression and acceleration techniques for the efficient execution of relevant deep learning models, and determine different probabilities of detection success. Afterwards, we integrate these success indicators in the localization layer of nodes and by information diffusion we verify increased global location awareness against missing detections. A variety of simulated experiments performed in KITTI dataset, as well as CARLA simulator, indicate that the localization layer is able to address highly efficient the missing detections for each accelerated model, effectively realizing the vision of distributed swarm intelligence.