前驱差速轮式农业机器人的BP定位估计方法

BP Neural Network Based Localization for a Front- Wheel Drive and Differential Steering Agricultural Robot

针对一款自主研发的轮式农业移动机器人,提出了一种基于BP神经网络的轮式农业移动机器人定位估计方法。该机器人为前驱模式,依靠差速完成转向。定位方法是在前期的神经网络训练过程中,以左轮、右轮的驱动脉冲数为输入,以左右轮行进长度为输出,训练函数采用Bayesian规则;在后期的定位推算中,则假设机器人左右轮行进轨迹为两段同心圆弧,进而计算机器人几何中心位移。在与最小二乘定位估计方法的实验对比中,基于BP神经网络的定位估计方法具有更高的定位精度和更小的估计方差。

This paper presents a new BP-neural-network-based localization algorithm for a wheeled agricultural mobile robot, which is front- wheel drive and differential steering. Training the BP neural network is the first step of the localization algorithm. During this process, the drive pulse number is regarded as the input; the length of the left or right wheel＇s trajectory is regarded as the output; and Bayesian rule is used to generate the training function. Inducing the displacement of the robot＇s geometric center is the second step, in which the trajectories of the left and right wheels are assumed as two concentric circular arces. In the contrast experiments with the least square localization algorithm, the new proposed BP-neural-network based algorithm shows high accuracy and feasibility.