Topology and weather effects on CARLA autopilot performance in static driving simulation

Abstract Rule-based controllers are a critical component of autonomous driving system testing and have garnered significant attention in engineering practices aimed at achieving high reliability owing to their high logical interpretability. Therefore, the quantitative assessment of the robustness of rule-based controllers under various environmental conditions holds important practical significance for enhancing the reliability of autonomous driving systems. This paper presents a systematic benchmarking study aimed at evaluating the performance of the CARLA autopilot in static environments under varying weather conditions and urban topologies. We employ a rigorous experimental framework to decouple meteorological variations from urban topological structures, thereby isolating their respective impacts on the controller’s performance. This study conducted 100 independent simulation runs using five different weather conditions and four urban topologies of varying complexity from the baseline. The results indicate that urban topology is the primary determinant of the CARLA Autopilot’s performance within this static benchmark, whereas weather conditions show no significant effect on the CARLA Autopilot under the tested conditions. Furthermore, in the topologically complex town scenario, the controller exhibited lower vehicle speeds while maintaining a zero-collision record, reflecting a conservative strategy where speed is reduced under complex road conditions in the CARLA Autopilot. This study provides a refined benchmark for the performance of the CARLA Autopilot and highlights the necessity of adopting topology-aware control strategies in future autonomous driving system development. All experimental configurations have been documented to support reproducible research.