Uncertainty-Aware Bi-Level Trajectory Planning for Autonomous Vehicles in Underground Mine Roadways

Artikel i vetenskaplig tidskrift, 2026

In underground mines, the absence of positioning signals and the presence of heavy fog and dust lead to significant perception uncertainty. Meanwhile, the uncertainty caused by control errors that may lead to collisions cannot be ignored. Perception and control uncertainties result in significant uncertainties in the vehicle's pose (position and heading angle). In wide scenarios, pose uncertainty is handled by simply inflating the vehicle's bounding box with a conservative margin. However, in narrow underground mine roadways, over-inflation leads to no feasible solutions in trajectory planning, resulting in a low planning success rate (PSR). Conversely, under-inflation carries a risk of collision for autonomous vehicles, resulting in a low collision-free rate (CFR). In view of this, this paper proposes a bi-level trajectory planning algorithm that accounts for vehicle pose uncertainty. The first level employs a soft constraint to keep the vehicle away from obstacles and road boundaries, producing a safer trajectory without relying on fixed-margin inflation, thereby contributing to a high PSR. The second level applies uncertainty propagation theory to predict the pose distribution at each point along the trajectory generated by the first level. It then assesses whether the trajectory meets the required safety confidence level, which ensures a high CFR. This bi-level design avoids fixed-margin inflation, achieving both a high PSR and a high CFR. Finally, simulation experiments and scaled-down experiments in the real world are conducted. The results demonstrate that, when pose uncertainty is accounted for, the proposed algorithm improves the trajectory CFR by approximately 9% compared to the State-Of-The-Art (SOTA) method, without compromising PSR.