Research On Laser SLAM Algorithm Of Power Patrol Robot

With the growth of power consumption in society,the requirements for power quality have become higher,which means higher requirements for the quality of power inspections.Due to the traditional manual inspection method is easy to make wrong inspection and missed inspection,the reliability of the inspection quality cannot be guaranteed.To some extent,the use of electric inspection robots instead of manual inspections can improve the inspection quality and efficiency under the promotion of unattended substation and intelligent substation mode.Simultaneous Localization and Mapping(SLAM)method,as one of the key technologies of power inspection robots,plays a significant role in realizing power inspection operations.At present,due to the problems of pool real-time performance,large positioning accuracy errors,and low mapping quality in SLAM method,people will face various challenges when using electric inspection robot.For example,due to the positioning accuracy and low map quality,the robot is easy to encounter dangerous facilities in the substation,which affects the safety of the power station.Therefore,the accuracy and real-time performance of SLAM technology are required.This paper studies the laser SLAM algorithm of electric power inspection robot,the main research contents are as follows: In view of the problems of the existing particle filter SLAM algorithm,such as large computation and low mapping quality,firstly,as to estimate the probability distribution of robot pose with fewer particles,the thesis proposed to improve the proposed distribution by integrating the depth camera and lidar data in the prediction stage,which can reduce the number of particles in the resampling stage.Secondly,in the lidar point cloud registration stage,a grouped stepwise threshold judgment method is proposed to effectively reduce calculations of the Iterative Closest Point(ICP).In view of the problems of large trajectory deviation and low positioning accuracy of the Graph-Based SLAM,firstly,the thesis proposed the IMU motion compensation mechanism at the front end as to remove the point cloud distortion.And the feature extraction of lidar point cloud is proposed to improve the matching speed of point cloud while ensuring the matching accuracy.Secondly,in view of the low search efficiency in the loop closure,the branch and bound method is proposed to improve the search efficiency.Finally,in view of the low initial value accuracy of robot pose in back end optimization,the thesis proposed to adopt Unscented Kalman Filter to fuse the lidar point cloud matching information and the IMU pre-integration results to provide initial values for the back-end optimization,as to speed up the back-end iteration and improve positioning accuracy.Moreover,IMU pre-integral results,closure loop results and point cloud registration are added to the back-end optimization as constraints to improve map consistency.In the experiment,the public data set and robot platform are used to test the performance of the improved SLAM method based on particle filter and the improved SLAM method based on graph optimization.Experimental results show that the above two methods effectively improve the positioning accuracy and mapping quality of the inspection robot,reduce the robot’s trajectory drift,and have good real-time performance.