深度神经网络在AGV实时导航优化的应用研究  

Application of Deep Neural Network in AGV Real Time Navigation Optimization

作　　者：丁昌荣 金涛 李志军 周亮 DING Chang-rong;JIN Tao;LI Zhi-jun;ZHOU Liang(Yunnan Power Grid Co.,Ltd.,Yunnan Kunming 650000,China)

机构地区：[1]云南电网有限责任公司,云南昆明650000

出　　处：《机械设计与制造》2024年第8期128-134,共7页Machinery Design & Manufacture

基　　金：中国南方电网有限公司科技项目—基于AGV的物资智能化仓储关键技术研究与应用(050100KK5219001)。

摘　　要：为提高自动导航车辆(AGV)导航控制的自主性和智能性,提出了一种基于深度神经网络(DNNS)的混合智能实时最优控制方法。将AGV轨迹运动规划的避障问题转换为非线性最优控制问题(OCP),采用平滑变换来处理路径约束,通过高斯伪谱(GPM)法来获得问题求解路径规划的最优解。在此基础上设计基于DNNS的实时最优控制器,实现AGV的最优轨迹规划。数值实验结果表明,所设计的基于DNNS的最优控制方法能够生成最优控制指令,引导AGV到达目标位置,且对初始条件具有较高的鲁棒性,能够满足不同障碍的约束条件,实现自动导引车的实时导航优化策略。In order to improve the autonomy and intelligence of AGV navigation control,a hybrid intelligent real-time optimal control method based on deep neural network(DNNS)is proposed.The obstacle avoidance problem of AGV trajectory planning is transformed into a nonlinear optimal control problem(OCP).The smooth transformation is used to deal with the path constraints,and the Gaussian pseudo spectral(GPM)method is used to obtain the optimal solution of the problem.On this basis,a real-time optimal controller based on DNNS is designed to realize the optimal trajectory planning of AGV.The results of numerical experiments show that the optimal control method based on DNNS can generate the optimal control instructions to guide AGV to the target position,and has high robustness to the initial conditions.It can meet the constraints of different obstacles,and realize the real-time navigation optimization strategy of AGV.

关 键 词：深度学习 神经网络 AGV 非线性最优控制 轨迹运动规划 

分 类 号：TH16[机械工程—机械制造及自动化] TP183[自动化与计算机技术—控制理论与控制工程]