AGV Positioning System Based On XGBoost Model Multi-Sensor Data Fusion

With the optimization of AGV performance and the expansion of functions,the application field of AGV is constantly expanding and the complexity of applications is constantly increasing,which puts forward higher requirements for the design of AGV systems.The key technologies in the AGV system are positioning and navigation,path planning,obstacle avoidance,and trajectory tracking,and positioning technology is a prerequisite for ensuring AGV motion accuracy.A single sensor can no longer meet the robot's perception and positioning requirements in an increasingly complex environment,and multi-sensor fusion can learn from each other's strengths and weaknesses and give full play to their respective advantages,becoming the main means of AGV positioning.With the development of artificial intelligence,especially deep learning methods,the application of mobile robots is becoming more and more deep.An AGV localization method based on multi-sensor data fusion of XGBoost model was proposed,and an experimental platform was established for verification.The main work is as follows:1.Kinematics modeling of AGV.The system model of AGV with incomplete constraints was established,and an autonomous guided vehicle kinematics model were established to provide support for the study of positioning and navigation algorithms.2.Research the single-sensor AGV positioning method.Through the study of the single-sensor positioning algorithm of AGV by modeling each sensor separately,the inertial navigation positioning model and the laser SLAM model were mainly studied,and analyzed the influence of the above sensors on the AGV trajectory tracking accuracy,which lay the foundation for multi-sensor data fusion.3.Research on XGBoost multi-sensor data fusion.The XGBoost model was further studied and an improved XGBoost multi-sensor data fusion model was proposed,achieved two-dimensional motion plane AGV positioning data fusion.In addition,according to the influence of abnormal data on the accuracy of the model,the quartile based amplitude change limited sliding filtering algorithm was proposed,and the loss function of the XGBoost model was improved,so the positioning accuracy of the fusion algorithm was improved.4.Set up an experimental platform for real-world experiments.Based on the method proposed in this paper,an experimental platform was constructed,and accuracy experiments and robust experiments were designed under real scenarios to prove its effectiveness and accuracy.