Machine-Learning-Enabled Cooperative Perception for Connected Autonomous Vehicles: Challenges and Opportunities

Abstract

Connected and autonomous vehicles is a disruptive technology that has the potential to transform the current transportation system by reducing traffic accidents and enhancing driving safety. One major challenge of building such a system is how to realize effective and efficient cooperative perception among vehicles, which enables them to share local (raw or processed) perception data with each other or roadside infrastructures through wireless communications. As machine learning techniques become prevalent in autonomous vehicles, particularly in their perception subsystem, we articulate the possibility to design a machine-learning-enabled cooperative perception system for connected autonomous vehicles. Not only are the research challenges in designing cooperative perception presented, but we also focus on how to reduce communication and data processing latency in order to meet the stringent time requirements posed by autonomous driving applications. The article outlines the research challenges and opportunities in designing cooperative perception for autonomous vehicles, leveraging the recent research outcomes from machine learning, feature map quantification, millimeter-wave communications, and vehicular edge computing.