Logo: to the web site of Uppsala University

*In the curved navigation of a wall-climbing robot, a laser navigation calibration instrument is designed to help the robot position on the wall. Human error can interfere with the input data in navigation, resulting in the decline of the output data's accuracy. In this paper, we analyze the sensitivity index of human errors in the process of navigation. There are several methods in the literature to determine the sensitivity indices of various human errors. Researchers have provided its validity. Compared with the Nonparametric Spearman rank-order correlation method, the simple analysis of variance technique, and the connection weight method, the Mean Impact Value (MIV) algorithm allows the effect of the output variables corresponding to each perturbation in the input variable to be recorded. As a machine learning method widely used in data analysis, BP neural network can significantly improve the experimental efficiency. The paper applied a technique to study the sensitivity index of human errors in navigation. This method integrates the Mean Impact Value (MIV) algorithm with BP neural network model by MATLAB. In the experiment, one thousand arrays of data are generated according to the paper of Design of a Laser-based Calibration instrument for Robot's Location Positioning on A Curved Surface. And these one thousand arrays of data are used to train a BP neural network model by MATLAB. The result of the BP neural network model is reliable, with the whole R is 0.99341. Due to the perturbations caused by each human error, five hundred arrays of data are generated in the input variable. This sensitivity analysis method could obtain an array of mean impact variables of human error by the MIV algorithm, which corresponds to each perturbation in the input variable. The results indicate that the perturbations caused by human error in the laser rotation angle a are greater than those in the laser-assisted navigation calibration instrument's original coordinate position. And the output variables increase linearly with the increase of the input error.