基于深度学习的矿用救援机器人自动避障方法  

Automatic Obstacle Avoidance Method of Mining Rescue Robot Based on Deep Learning

作　　者：李骁龙 LI Xiaolong(Shuozhou Emergency Rescue Team,Shuozhou,Shanxi 036025,China)

机构地区：[1]朔州市应急救援队,山西朔州036025

出　　处：《自动化应用》2024年第3期15-18,共4页Automation Application

摘　　要：救援机器人自动避障性能存在不足,在实际中救援成功率和避障平滑系数较低,无法达到预期的避障效果,为此,提出基于深度学习的矿用救援机器人自动避障方法。首先,利用电子罗盘和超声波传感器感知救援机器人与障碍物的方位角和距离,搭建救援机器人空间状态;然后,建立具有三层卷积层和两层全连接层结构的深度学习网络模型,并搭建用于深度学习网络模型训练的救援机器人避障动作集合;最后,通过深度学习网络模型训练救援机器人的空间状态信息,提取救援机器人移动的空间特征,自动生成避障决策。实践证明,应用该设计方法后,救援机器人避障成功率在95%以上,平滑系数在0.85以上,具有良好的应用前景。There are shortcomings in the automatic obstacle avoidance performance of rescue robots.In practice,the success rate of rescue and the obstacle avoidance smoothness coefficient are low,which cannot achieve the expected obstacle avoidance effect.Therefore,a deep learning based automatic obstacle avoidance method for mining rescue robots is proposed.Firstly,use an electronic compass and ultrasonic sensors to sense the azimuth and distance between the rescue robot and obstacles,and construct the spatial state of the rescue robot.Then,a deep learning network model with three convolutional layers and two fully connected layers is established,and a set of obstacle avoidance actions for rescue robots is constructed for training the deep learning network model.Finally,the spatial state information of the rescue robot is trained through the deep learning network model,and the spatial features of the rescue robot′s movement are extracted to achieve automatic generation of obstacle avoidance decisions.Through experiments,it has been proven that after using the design method,the success rate of obstacle avoidance for rescue robots is above 95%,and the smoothness coefficient is above 0.85,which has good application prospects.

关 键 词：深度学习 救援机器人 自动避障 电子罗盘 超声波传感器 空间状态 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]