Priority Optimization for Autonomous Driving Systems to Meet End-to-End Latency Constraints

In autonomous driving (AD) systems, complex data dependencies exist between tasks with different activation rates, making it very hard to analyze the system’s timing behaviors. This paper formulates an AD system as a multi-rate directed acyclic graph (DAG) and introduces a novel reaction time bound for critical chains within this multi-rate DAG. Furthermore, we introduce a priority assignment strategy tailored to optimize priority allocation, effectively minimizing the reaction time of critical task chains. This strategy comes with theoretical guarantees, ensuring that the achieved latency bound is only slightly higher than the ideal one. Our empirical work demonstrates that the newly proposed reaction time bound outperforms current standards, achieving an average improvement of $5.46 \%$. Furthermore, our strategy for priority assignment significantly enhances the success rate of achieving timing correctness in the AD system, exceeding the baseline method by a notable $19.24 \%$.