无线电工程

针对现有巡检机器人导航避障存在的不足,将深度学习技术与路径规划相结合,提出了一种融合场景理解与A*寻路算法的巡检机器人避障方法。该方法采用基于编码-解码结构的深层卷积神经网络构建高精度场景理解网络,获取巡检机器人道路场景信息。利用中值滤波、最大连通域和C空间变换等操作,提取出场景信息中机器人可行道路区域,并转化为二维栅格地图。通过基于栅格地图的A*路径规划算法搜索出最优避障路径,指导机器人完成避障动作。同时,考虑到实际道路场景的高重复性,引入特征差分结构来降低冗余计算,保障巡检机器人导航避障效率。实验结果表明,所提方法在场景理解以及避障路径规划的精度和计算效率方面都得到了有效的平衡,并能适应不同场景,鲁棒性较高。同时,在真实变电站道路环境中,该方法也能高效获取场景信息,并准确指导巡检机器人实现实时导航避障。

To solve the problem of the shortcomings of existing inspection robots for navigation and obstacle avoidance, an inspection robot obstacle avoidance method integrating scene understanding and A* path finding algorithm is proposed by combining deep learning technology with path planning. Firstly,the deep convolutional neural network based on encoding-decoding structure is used to construct a high-precision scene understanding network to obtain the road scene information of inspection robot. Secondly,the feasible road area of the robot is extracted from the scene information by using the operations of median filter,maximum connected domain and C-space transformation,and transformed into a 2D grid map. Finally, the optimal obstacle avoidance path is searched by A * path planning algorithm based on grid map to guide the robot to complete the obstacle avoidance action. At the same time,considering the high repeatability of the actual road scene,the characteristic difference structure is introduced to reduce the redundant calculation and ensure the navigation and obstacle avoidance efficiency of the inspection robot. The experimental results show that the proposed method has an effective balance in the accuracy and computational efficiency of scene understanding and obstacle avoidance path planning,and can adapt to different scenes with high robustness. At the same time,in the real substation road environment, this method can also efficiently obtain scene information and accurately guide the inspection robot to realize real-time navigation and obstacle avoidance.