Parallel Perception of Autonomous Vehicle in Degraded Coal Mines

The working environment in coal mines is complex and constantly changing. Harsh conditions, such as narrow passageways, irregular road shapes, and unpredictable obstacles, pose significant challenges for the precise perception of unmanned mining vehicles. To address these issues, this paper proposes a positioning method that tightly couples an Inertial Measurement Unit (IMU) with LiDAR to overcome the challenge of precise positioning in degraded underground spaces where a single LiDAR alone cannot achieve accurate localization.First, the LiDAR point cloud is segmented, and IMU pre-integrated poses are used to eliminate nonlinear motion distortions. Line and surface features are extracted from the obtained point cloud. Next, the line and surface features of adjacent frames are matched, and in the layered pose estimation process, IMU pre-integrated pose initial values are fused to reduce the number of computational iterations, improve feature point matching accuracy, and compute the current frame's pose. Finally, local map factors, IMU factors, and keyframe factors are inserted into the factor graph to optimize the pose. Keyframes are matched with local maps, and map construction is achieved through an octree structure.Considering the specific limitations of coal mining operations in the real environment, which are characterized by high experimental difficulty, cost, limited and hard-to-obtain data, this paper combines the ACP method to simulate various coal mine tunnel scenarios within an artificial system. It includes features such as uneven ground, curves, slopes, and other complex parameters to experimentally validate the proposed method. This approach not only reduces experimental risk and costs but also significantly enriches the experimental data.The experimental results demonstrate that the method presented in this paper exhibits strong robustness and accuracy in various scenarios. The positioning accuracy of unmanned mining vehicles in actual scenarios is improved by 74% to 91% compared to traditional methods.