Task Scheduling for Autonomous Vehicles with Heterogeneous Processing via Integration of the CARLA Simulator and StarPU Runtime
E. L. R. da Silva, Henrique Cota de Freitas · 2025
Autonomous driving vehicles (ADV) rely on complex sensor systems and real-time decision making algorithms to function effectively. Efficient task scheduling is crucial for optimizing heterogeneous computing resources, which ADVs require to process data in real-time. This paper presents a novel approach that integrates the CARLA simulator, a widely used platform for testing and validating autonomous systems, with the StarPU runtime system that offers flexible task scheduling for heterogeneous hardware. By combining these two systems, the paper explores different scheduling strategies for ADV perception and localization tasks. The methodology combines CARLA's simulation capabilities and StarPU's dynamic resource allocation to evaluate task execution times. The results show that this integration provides a practical platform for testing and evaluating scheduling strategies in a realistic simulation environment, significantly enhancing ADV performance in sensor processing and task management.