Abstract: This paper develops a UAV-assisted vehicle–roadside unit (RSU) collaborative edge computing system for distributed monitoring tasks in heavy-haul trains. The model jointly considers the limited computing capability of onboard devices, the time-varying shared workload of RSUs, and the energy-constrained computing capability of UAVs. The monitoring data upload volume, wireless transmit power, task offloading ratio, and edge computing resource allocation are jointly optimized to minimize the task timeout failure rate. To address the high-dimensional, dynamic, and non-convex nature of the problem, it is formulated as a Markov decision process, and a dynamic resource scheduling method based on multi-agent proximal policy optimization is proposed. A centralized-training and decentralized-execution framework is adopted to coordinate task offloading and resource allocation according to task, channel, queue, and available resource states. Simulation results show that the proposed method effectively reduces the task timeout failure rate and average processing delay, thereby improving the real-time performance and reliability of heavy-haul train condition monitoring.