Designing and Implementing AoI-Optimized Scheduling for Autonomous Driving Systems

Chen Tang, Tuo Shi, Qian Xu, Kui Wu, Nan Guan, Jen-Ming Wu, Jianping Wang

Abstract: Autonomous Driving Systems (ADS) incorporate complex algorithm stacks, including sensing, localization, perception, prediction, planning, and control. To enhance the safety and comfort of ADS, it is crucial to utilize the most recent sensor data and meticulously schedule these algorithm stacks for better system communication. Since the data-flow communication within ADS leverages sensor data for final planning decisions, our objective is to minimize the Age of Information (AoI), a metric that assesses the freshness of the processed information. The optimization of AoI on ADS, however, poses significant challenges due to computing resource limitations and the complicated nature of ADS operations. Moreover, most ADS platforms, such as Autoware, are built on Robot Operating System 2 (ROS 2), whose special execution behavior introduces additional complexities in scheduling algorithm design. To tackle these challenges, we propose an AoI-based scheduling optimization framework specifically for ADS. This involves developing effective algorithms to streamline and enhance the scheduling processes, as well as customizing ROS 2 at the system level to optimize AoI further. Experimental evaluations of our proposed policy against existing state-of-the-art scheduling policies in ROS 2 demonstrate notable improvements in the efficiency and responsiveness of ADS.

External IDs:dblp:conf/infocom/TangSX0GWW25