REAL-TIME PERFORMANCE OPTIMIZATION FOR AUTONOMOUS DRIVING SYSTEMS BASED ON EDGE COMPUTING ARCHITECTURE

Abstract

Autonomous driving technology is rapidly evolving as a cornerstone of next-generation intelligent transportation systems. However, achieving real-time responsiveness and operational reliability remains a significant challenge due to the latency and bandwidth limitations inherent in traditional cloud-centric architectures. To address this issue, this study proposes a novel edge computing-based optimization framework that restructures the computational hierarchy across terminal, edge, and cloud layers to support time-critical decision-making in autonomous vehicles.The methodology integrates hierarchical task offloading, modular container deployment using Docker, and dynamic resource orchestration via Kubernetes. Perception and decision-making modules were optimized with lightweight deep learning models and heuristic planning algorithms, and the entire system was validated using the Baidu Apollo platform under realistic urban and highway scenarios.Experimental evaluation demonstrates that the proposed architecture achieves a 23.6% reduction in end-to-end latency and maintains object detection performance with only a 2.8% decline in mean Average Precision (mAP) when using compressed models. Furthermore, decision-making accuracy improved significantly with the enhanced RRT algorithm, which reduced planning errors by 66.7% compared to traditional methods. Bandwidth efficiency and scalability were also improved under varying workloads.In conclusion, the edge computing-based framework effectively mitigates the latency bottlenecks associated with centralized processing while preserving model accuracy and improving decision precision. This work lays a practical foundation for scalable, low-latency deployment of autonomous driving systems and highlights the transformative role of edge computing in advancing real-time vehicular intelligence. Future research should explore adaptive resource allocation strategies and robust security mechanisms to enhance performance under high-concurrency and adversarial conditions..