基于深度学习的巡检机器人避障轨迹自动控制系统设计

Design of Automatic Control System for Obstacle Avoidance Trajectory of Inspection Robot Based on Deep Learning

摘要:

为了提高巡检机器人在复杂环境下的避障能力,使机器人能够安全地完成巡检任务,设计基于深度学习的巡检机器人避障轨迹自动控制系统。设计由CCD传感器、信号处理芯片等设备组成的工业智能视觉CCD相机,基于FPGA和USB2.0的视频采集卡传输采集数据,完成硬件部分的设计。在软件设计中,对采集的图像实施目标分割、双目目标匹配等预处理,通过对摄像头实施双目视觉标定获取障碍物空间位置三维信息,基于深度学习中的CRNN设计机器人自主避障规划网络模型,并设计模糊轨迹控制器,实现避障中的轨迹自动控制。系统测试结果表明,设计系统最终成功避开了三个动态障碍物,最大轨迹控制误差的最大值为1.45°,最小轨迹控制误差的最大值为0.62°,动态避障巡检速度始终在3.5m/s左右,表现出了精准而稳定的轨迹控制效果。

Abstract:

In order to improve the obstacle avoidance ability of inspection robots in complex environments and enable them to safely complete inspection tasks, an automatic control system for obstacle avoidance trajectory of inspection robots based on deep learning is designed. Design an industrial intelligent visual CCD camera composed of CCD sensors, signal processing chips, and other devices. Based on FPGA and USB2.0 video capture cards, transmit and collect data, and complete the hardware design. In software design, preprocessing such as target segmentation and binocular target matching are implemented on the collected images. The three-dimensional information of obstacle spatial position is obtained through binocular visual calibration of the camera. Based on CRNN in deep learning, a robot autonomous obstacle avoidance planning network model is designed, and a fuzzy trajectory controller is designed to achieve automatic trajectory control in obstacle avoidance. The system testing results show that the designed system successfully avoided three dynamic obstacles, with a maximum trajectory control error of 1.45° and a maximum minimum trajectory control error of 0.62°. The dynamic obstacle avoidance inspection speed is always around 3.5m/s, demonstrating precise and stable trajectory control effects.