A Lightweight Collaborative Resource Scheduling Method for Autonomous Driving in Resource-Heterogeneous Environments

Aiming at the problems of strong resource heterogeneity and inefficient in task-resource matching in autonomous driving edge computing, this paper proposes a Lightweight Collaborative Resource Scheduling Algorithm (RHLCA). The algorithm adopts a dynamic priority assignment strategy and constructs a dual-core scheduling architecture with a closed-loop optimization process. Through an improved auction model, it intelligently allocates computing resources by integrating task priority, resource requirements, node load, and energy consumption. A lightweight Deep Q-Network (DQN) is leveraged to dynamically allocate bandwidth, ensuring system fairness while guaranteeing low latency for high-priority tasks. Furthermore, the algorithm establishes a closed-loop mechanism of “decision-scheduling-feedback”, which dynamically adjusts priority weights based on real-time performance feedback to achieve global optimization and lightweight adaptive deployment. Experimental results demonstrate that the RHLCA improves the task completion rate by 7.68%, 12.02%, and 18.27% compared to the DRL, GA, and FIFO algorithms, respectively, effectively enhancing the real-time performance and reliability of autonomous driving tasks in heterogeneous environments.