结合飞行、爬行两种运动模式的混合式架空输电线路巡检机器人相比于传统的攀爬式机器人和无人机在巡检精度、灵活性和稳定性上都有明显优势，但是机器人自主性不足限制了其推广应用。本项目拟结合机器人学、机器视觉、机器学习等学科理论和方法，针对混合式机器人自主巡检面临的环境识别、定位和运动规划等关键问题开展深入研究。首先基于激光雷达和视觉信息，进行输电走廊典型特征的识别研究，获取输电线、杆塔等典型对象信息；其次结合已知点云地图，通过子采样地图与图像特征匹配，实现巡线机器人的实时定位；然后基于生成对抗网络和模仿学习策略，进行巡检障碍感知和机器人运动规划研究，实现输电走廊动态时变环境下的全线路巡检最优路径规划；最后将上述成果应用于现有机器人平台，验证其有效性。通过本项目的实施，突破制约架空线自主巡检的关键技术瓶颈，为混合式巡检机器人的实用化和电网的智能化奠定基础。

Compared with the traditional climbing robot and UAV, the hybrid-mode inspection robot for overhead transmission line has obvious advantages over the inspection precision, flexibility and stability, but the lack of autonomy of the robot restricts its popularization and application. This project is based on the theory and methods of robotics, machine vision, machine learning and other disciplines, aiming at the key problems such as the environment identification, location and motion planning of autonomous inspection with the hybrid-mode robot . First, based on LiDAR and visual information, the identification of typical features of transmission corridors is carried out, and the typical object information of transmission lines and towers is obtained. Secondly, combining with the known point cloud Map, the real-time location of the inspection robot can be realized by matching the point cloud sub-sampling map and the image feature. Then based on the generation adversarial network and imitation learning strategy, the study of obstacle perception and robot motion planning is carried out to realize the.optimal path planning in the dynamic time-varying environment of transmission corridor. Finally the results are applied to the existing robot platform to verify their validity. Through the implementation of this project, the key technical bottleneck restricting the independent inspection of overhead lines can be broken, which lays the foundation for the practicality of the hybrid-mode inspection robot and the intelligence of the power grid.