Design And Implementation Of Localization Method For Mobile Robot Based On Multi-sensor Fusion

Indoor mobile robots have been widely used in families,hospitals,restaurants,factories and other environments.At present,the mobile robot is to complete relatively simple work,with the increase of application scenarios and the emergence of new requirements,it is necessary to improve the intelligent and autonomous movement ability of mobile robots.The localization ability of mobile robot is one of the important technologies to realize the intelligent and autonomous degree of mobile robot.In this thesis,by studying the domestic and foreign mobile robot localization related literature,the main indoor localization methods are investigated,and the in-depth research is carried out from two aspects of probabilistic localization and recurrent neural network localization,the main contents of the study are as follows:(1)The overall architecture of localization algorithm is designed,including dynamic particle number particle filter localization and recurrent neural network localization.According to the design,the hardware and software platform of the positioning system is built,and the indoor localization system is arranged in the experimental environment,and the location and pose calculated by the system is taken as the true value of localization.(2)The Kalman filter algorithm is studied and analyzed.Based on the location data of odometer and IMU,the Kalman state transfer equation and measurement update equation of the localization system are established,then,according to the theoretical analysis and experimental test,the value of each parameter in the system is determined,experimental results show that Kalman filter can provide local accurate localization information.(3)Combined with localization system,the process of improving particle filter is studied,aiming at the contradiction between reducing calculation time and improving localization accuracy of fixed particle number,according to the motion state of robot,the method of generating the number of particles dynamically by fuzzy control is proposed,and study the relationship between robot state and particle numbers through simulation experiment.Through the analysis of the experimental data,it is verified that the dynamic particle number particle filter can obtain better localization accuracy in turning scenes.(4)The calculation and analysis process of data and the correlation of historical data by recurrent neural network are studied,Long short-term memory(LSTM)recurrent neural network model for localization is designed,the original output of LSTM neural network is fully connected to get the final localization result.The experimental results show that LSTM neural network has a good localization effect in the original environment,can maintain the localization stability in a long time,and can deal with the interference of environmental changes.