Laser-based Mobile Robot Relocalization Research

With the development of the times,mobile robots have become more and more widely used in society.The autonomous capability of the robot is one of the important indicators to evaluate the performance of the robot.In order to achieve the function of relocalization,this paper focuses on there bullets points:map construction,mobile robot relocalization based on probabilistic model and mobile robot relocalization based on deep learning.Among them,the construction of a priori map mainly provides map information for mobile robots,the relocalization method based on deep learning is used to predict the approximate position of the robot in the environment,and the probabilistic model is used to carry out the pose accuracy of the robot.In the research of simultaneous localization and mapping,this paper comparatively studied Rao-Blackwelliezed based particle filter mapping algorithm and Google's open source Cartographer mapping algorithm.Among them,Rao-Blackwelliezed based particle filter mapping algorithm has low computational,complexity and good real-time performance on the robot platform.Cartographer mapping algorithm is better in the point of closed-loop optimization,so the accuracy is higher.Therefore,in this paper,Cartographer mapping algorithm was selected to provide a priori map information for the subsequent research of mobile robots.This paper firstly proposes an improved particle filtering algorithm,which uses the proposed image matching algorithm to predict the approximate position of the robot,and then uses the particle filtering algorithm to optimize the precise pose of the robot.In addition,this paper proposes an indoor mobile robot relocalization strategy based on the combination of deep learning pose prediction and local particle filtering algorithm.The strategy consists of a two-dimensional laser scan preprocessing module and a neural network based pose predicted module and a local particle filtering optimization module.The strategy can predict the robot global position when the mobile robot gets on line,and then use the local particle filter algorithm to optimize the accuracy of the predicted position.In addition,this paper also extends the indoor robot relocalization strategy to a large outdoor open environment,and proposes an outdoor mobile robot relocalization algorithm.The algorithm includes a preprocessing module for outdoor 3D laser data and a neural network pose regression module.The large-scale open environment changing robot relocalization strategy proposed in thispaper can efficiently process 3D laser data and accurately predict the global pose of the outdoor robot based on the network model combining VGG and Dense-block.In this paper,the indoor and outdoor robots carrying 2D and 3D lasers are used as hardware platforms,and multiple real experimental scenarios(including indoor similar environment and outdoor open environment)are selected to verify the effectiveness of the mobile robot relocation strategy proposed in this paper.