Cooperative Right-of-Way Assignment for Autonomous Mining Trucks at Multiple Open-Pit Mine Intersections

This research presents a cooperative approach to assigning right-of-way to autonomous mining trucks, aiming to optimize passage order across multiple intersections and enhance transportation efficiency in open-pit mines. Unlike traditional methods that focus on managing a single or limited number of intersections, Multi-Agent Reinforcement Learning provides a robust framework capable of handling complex coordination across multiple intersections, thus improving collaborative management. MARL leverages its dynamic, decentralized control to provide significant advantages in mining operations. In contrast to urban road environments, open-pit mines face challenges such as complex spatial interdependencies between intersections and the need for synchronized actions among trucks. However, open-pit mines benefit from full observability and predictable traffic flows due to their closed nature and fixed haulage tasks. To address these challenges, this research proposes a globally aware and regionally cooperative strategy integrated within the Independent Advantage Actor-Critic framework. The Graph Convolutional Network captures the spatial relationships across all intersections, while a Spatial-Temporal Graph Attention Network analyzes the intricate interdependencies among intersections. Additionally, a spatial discount factor is developed to modulate rewards across neighboring intersections. Simulations and real-world applications at a large-scale mining site in Inner Mongolia, China validate the efficiency of the proposed cooperative right-of-way assignment method.