Mobile Robot Localization Algorithm Based On Multi-sensor Fusion And Point Cloud Matching

With the new trend of artificial intelligence,artificial intelligence technology has developed rapidly.As an important carrier of artificial intelligence,mobile robots technology develops and innovates with an unprecedented speed.When mobile robots performing tasks on an industrial site,the most critical issue for mobile robots is how to get to the target position.This involves two problems: navigation and obstacle avoidance,and the accurate localization is the prerequisite to solve the problems.A mobile robot should be equipped with sensors to infer their pose in the environment.With the expansion of mobile robots application,the requirement of the localization accuracy is getting higher.Generally speaking,it is difficult to ensure the localization accuracy by the single sensor data.Therefore,the multi-sensor data fusion become a mainstream solution in the current mobile robots localization field.In this thesis,from the perspective of the multi-sensor fusion and the point cloud matching,the mobile robot localization algorithm is carried out.By proposing the localization algorithm of odometry and inertial measurement unit data fusion based on the unscented Kalman filtering and the normal distributions transform algorithm which is based on the global normal distributions map,the accuracy of the global pose is improved.By the way,the problems of the low localization accuracy and the high localization time of the mobile robot are solved.The main contents include:(1)Proposing a localization algorithm by the odometry and inertial measurement unit data fusion based on the unscented Kalman filtering.In this thesis,a robot motion model is established,and the model is used to predict the mobile robot pose.Furthermore,the odometry and inertial measurement unit are used as observations in the unscented Kalman filtering to correct the predicted pose.At the same time,the noise is corrected according to the measurement time difference between the odometry and inertial measurement unit.The experiments show that the proposed method can greatly reduce the measurement error of the odometry and inertial measurement unit,and improve the accuracy of the localization.(2)Developing an adaptive Monte Carlo localization algorithm with the fusion pose of the odometry,inertial measurement unit and laser's measurement.By sampling data from the difference of the two fusion pose,the accumulated errors in the odometry and inertial measurement unit are prevented from the Monte Carlo localization algorithm,and a stable multi-sensor fusion pose is achieved.The experiments prove that the algorithm can solve the cumulative error in the fusion pose of odometry and inertial measurement unit,and realize the tracking of the global pose of the mobile robot.(3)Proposing a normal distributions transform pose correction algorithm based on the normal distributions map.The multi-sensor fusion pose is transformed into a transformation matrix as the initial transformation of the normal distributions transform algorithm.Furthermore,both the global grid map and the laser point cloud are converted into the normal distributions,which is matched by the normal distributions transformation algorithm.By the way,the multi-sensor fusion pose is corrected.The experiments show that the localization accuracy of the mobile robot can be improved by the proposed method in a short time.(4)The localization accuracy and real-time performance of the mobile robot are tested.In these two typical environments,indoor and outdoor,the performance of the multi-sensor fusion and the point cloud matching localization algorithms are evaluated with two indicators: repeated localization accuracy and average localization time.The experimental results show that the proposed method can achieve fast and accurate localization on the mobile robot in this thesis.