Abstract: With the sustained expansion of the power grid scale, the inventory of high-voltage electrical tools and instruments has witnessed a considerable increase. Conventional manual testing procedures are constrained by inherent limitations, including suboptimal operational efficiency, inconsistent assessment criteria, and elevated safety risks, rendering them inadequate for the high-frequency, standardized, and uninterrupted testing requirements imposed by the ongoing high-quality development of the power system. This study centers on the scheduling challenges associated with heavy-duty automated guided vehicle (AGV) clusters, which serve as the primary conveyance medium for tool and instrument transportation. By integrating the Double Deep Q-Network (DDQN) and Advantage Actor-Critic (A2C) reinforcement learning paradigms, a distributed multi-AGV collaborative scheduling framework is formulated to accommodate the dynamic characteristics of high-voltage testing workshops. Furthermore, a hybrid fault diagnosis and positioning error compensation mechanism is established to mitigate typical engineering obstacles, including AGV inter-vehicle collisions, system deadlocks, localization inaccuracies, and unscheduled equipment downtime.