目前，我国煤矿井下掘进工作面支护设备的自动化程度普遍较低，随着近年来智能化掘进设备的快速研发和投入应用，掘进工作面“掘快支慢”的问题日益突出，支护设备的工作效率已经成为制约煤矿行业发展的关键因素，提高支护设备的自动化水平是改善当前“采掘失衡”问题的有效方案。因此，以双臂钻锚机器人系统为研究对象，考虑煤矿井下实际工况环境，提出一种基于单激光束信息的钻锚机器人机身视觉定位方法和基于语义分割的煤矿井下钢带锚孔视觉定位方法，并建立钻锚机器人钻臂的运动学模型，提出一种受限空间下钻锚机器人钻臂的轨迹规划方法，实现钻锚机器人的自主定位与跟踪控制。

针对钻锚机器人机身定位问题，以煤矿井下掘进工作面固有激光指向仪特征为视觉定位合作标靶，提出了基于单激光束信息的钻锚机器人机身定位方法和激光指向仪点线特征提取方法，解决了现有视觉定位方法因标靶结构复杂而导致标定繁琐的问题。另外，分析激光指向仪光斑图像特征，针对其高能量的特点，提出了一种基于二维反正切函数拟合的激光光斑中心提取方法，得到了比传统基于椭圆拟合和高斯拟合的光斑中心提取方法更加精确的结果。

针对煤矿井下钢带锚孔定位问题，提出了一种基于语义分割的钢带及锚孔区域分割方法，以精度和效率更高的ConvNeXt模型为主干网络，引入卷积注意力模块（CBAM）提高网络对钢带锚孔这类小目标的特征提取精度，融合空洞卷积、条形池化和Unet密集连接的优点提出了深层特征融合模块（DFFB）和浅层特征融合模块（SFFB），提高了传统DeepLab V3+网络对深层语义信息边缘细节信息的捕获能力。另外，提出了一种基于深度相机的钢带锚孔视觉定位方法，该方法以钢带分割结果为掩码与对齐后的RGB图像和深度图像进行运算，并基于相机模型得到钢带区域的点云数据。进而基于RANSAC算法对钢带区域点云数据进行平面拟合，得到钢带平面的姿态信息。最后结合锚孔中心提取结果基于相机模型得到钢带锚孔的位置信息。

针对钻锚机器人钻臂控制问题，在机身和钢带锚孔定位数据的基础上，分析钻锚机器人钻臂结构，基于D-H法构建了钻臂正运动学模型，并基于牛顿-拉弗森数值方法计算钻臂逆解，求解钻锚机器人钻臂的可达工作空间和有效工作空间，得出了钻锚机器人钻臂处于巷道和机身构成的受限空间的研究前提。另外，构建钻锚机器人机身、钻臂和巷道的包围盒并进行钻臂碰撞检测和碰撞分析。针对RRT*算法在高维空间下存在目的性差和采样效率低的问题，结合钻锚机器人钻臂碰撞分析结果及其在受限空间下的工作特点，引入目标引力的思想，提出了一种基于改进RRT*的钻锚机器人钻臂路径规划方法。最后，基于LQR方法实现了钻锚机器人钻臂的跟踪控制。

最后，在实验室模拟巷道环境搭建实验平台对提出的基于单激光束信息的钻锚机器人机身视觉定位方法、基于语义分割的煤矿井下钢带锚孔视觉定位方法和受限空间下钻锚机器人钻臂的轨迹规划及跟踪控制策略进行了实验。结果表明，基于单激光束信息的钻锚机器人机身视觉定位方法在50 m的范围内平均误差小于58.64 mm；基于语义分割的煤矿井下钢带锚孔视觉定位方法具有较高特征分割精度，同时钢带锚孔平均定位误差小于4.563 mm；受限空间下钻锚机器人钻臂的轨迹规划及跟踪控制策略具有良好的规划和跟踪性能。

At present, the degree of automation of support equipment in underground excavation face of coal mines in China is generally low. With the rapid development and application of intelligent excavation equipment in recent years, the problem of "fast excavation and slow support" in excavation face has become increasingly prominent. The working efficiency of support equipment has become a key factor restricting the development of coal mine industry. Improving the automation level of support equipment is an effective solution to improve the current "unbalance between mining and heading" problem. Therefore, focusing on the dual-arm drilling and anchoring robot system, and considering the actual working conditions in underground coal mines, this study presents a methods for vision-based localization of a drilling and anchoring robot using a single laser beam, and a methods for vision-based localization of steel belt anchor holes in underground coal mine shafts using semantic segmentation. Additionally, a kinematic model of the drill arm of the drilling and anchoring robot is established, and a trajectory planning and tracking control strategies for the drilling arm of the drilling and anchoring robot in confined spaces is proposed to achieve autonomous localization and tracking control of the drilling and anchoring robot.

Addressing the positioning challenge of the drilling and anchoring robot body, this study utilizes the inherent features of the laser pointing device at the underground heading face of coal mines as a visual positioning collaborative target. We propose a method for the positioning of the drilling and anchoring robot body based on single laser beam information, along with a method for extracting point-line features of the laser pointing device. This approach effectively resolves the issue of tedious calibration in existing visual positioning methods due to the complex structure of the target. Additionally, through the analysis of the laser pointing device's spot image characteristics, we introduce a method for extracting the center of the laser spot based on two-dimensional arctangent function fitting, leveraging its high-energy characteristics. This method yields more precise results compared to traditional methods based on ellipse fitting and Gaussian fitting for laser spot center extraction.

For addressing the localization issue of steel belt anchor holes in underground coal mines, a segmentation method based on semantic segmentation is proposed. This method utilizes the ConvNeXt model, known for its higher accuracy and efficiency, as the backbone network. It incorporates the Convolutional Block Attention Module (CBAM) to enhance the feature extraction accuracy of the network for small targets such as steel belt anchor holes. Moreover, it introduces the advantages of dilated convolution, strip pooling, and Unet dense connections to propose the Deep Feature Fusion Block (DFFB) and Shallow Feature Fusion Block (SFFB), thereby improving the ability of the traditional DeepLab V3+ network to capture deep semantic information and edge details. Furthermore, a visual localization method for steel belt anchor holes based on depth cameras is proposed. This method performs computations using the segmentation result of the steel belt as a mask with aligned RGB images and depth images. It derives point cloud data of the steel belt region based on the camera model. Subsequently, the RANSAC algorithm is employed to fit planes to the point cloud data of the steel belt region, obtaining the pose information of the steel belt plane. Finally, the position information of the steel belt anchor holes is determined by combining the extraction results of anchor hole centers with the camera model.

Aiming at the control problem of the drill arm of the drilling and anchoring robot, based on the positioning data of the fuselage and the steel belt anchor hole, the structure of the drill arm of the drilling and anchoring robot is analyzed. The forward kinematics model of the drill arm is constructed based on the D-H method, and the inverse solution of the drill arm is calculated based on the Newton-Raphson numerical method. The reachable workspace and effective workspace of the drill arm of the drilling and anchoring robot are solved, and the research premise that the drill arm of the drilling and anchoring robot is in the restricted space composed of the roadway and the fuselage is obtained. In addition, the bounding box of the body, drilling arm and roadway of the drilling and anchoring robot is constructed, and the collision detection and collision analysis of the drilling arm are carried out. Aiming at the problem of poor purpose and low sampling efficiency of RRT* algorithm in high-dimensional space, combined with the analysis results of drill arm collision of drilling and anchoring robot and its working characteristics in restricted space, the idea of target gravity is introduced, and a path planning method of drill arm of drilling and anchoring robot based on improved RRT* is proposed. Finally, based on the LQR method, the tracking control of the drilling arm of the drilling and anchoring robot is realized.

Finally, experimental validation was conducted in a laboratory-simulated tunnel environment to evaluate the proposed methods for vision-based localization of a drilling and anchoring robot using a single laser beam, vision-based localization of steel belt anchor holes in underground coal mine shafts using semantic segmentation, and trajectory planning and tracking control strategies for the drilling arm of the drilling and anchoring robot in confined spaces. The results show that the average error of the visual positioning method based on single laser beam information is less than 58.64 mm in the range of 50 m. The visual positioning method of steel belt anchor hole in coal mine based on semantic segmentation has high feature segmentation accuracy, and the average positioning error of steel belt anchor hole is less than 4.563 mm. The trajectory planning and tracking control strategy of the drilling arm of the drilling and anchoring robot in the confined space has good planning and tracking performance.