Abstract: Bidirectional single-lane roadways are widely present in underground auxiliary transportation roadways of coal mines, where vehicles traveling in opposite directions must rely on passing bays for meeting and yielding, and multi-vehicle parallel operations are prone to lead to competition for spatiotemporal resources. In the absence of coordinated scheduling and conflict resolution, traffic congestion and delays in vehicle transportation tasks may occur and even lead to safety accidents. Existing studies have limitations in accurately modeling underground traffic conflicts and efficiently solving path optimization problems under complex constraints. To address these issues, a path optimization method for underground trackless rubber-tyred vehicles in coal mines based on an improved large neighborhood search algorithm was proposed. With the objective of minimizing travel distance cost, vehicle startup cost, and time window violation cost, traditional vehicle routing problem constraints including rated vehicle load, maximum travel distance, travel time, and demand point time windows were considered, and meeting rules based on passing bays as well as spatiotemporal conflict avoidance constraints were introduced to construct a path optimization model for underground trackless rubber-tyred vehicles. To obtain high-quality feasible solutions within an acceptable time, an Improved Adaptive Large Neighborhood Search (IALNS) algorithm was designed. This algorithm adopted a hierarchical mechanism with upper-level and lower-level paths to improve search efficiency, integrated multiple destroy and repair operators, introduced a simulated annealing criterion to balance the global exploration and local search capabilities, and applied a perturbation restart strategy based on an elite solution pool to avoid being trapped in local optima. Experimental results showed that the IALNS algorithm obtained optimal solutions consistent with those of the Gurobi solver while achieving higher efficiency and stability. Compared with genetic algorithms, particle swarm optimization algorithms, and adaptive large neighborhood search algorithms, the IALNS algorithm demonstrated improvements in solution quality, computational speed, and stability. Under different passing bay density conditions, the IALNS algorithm maintained stable solution performance.