Research On SLAM Algorithm For Mobile Robot Based On Graph Optimization

Simultaneous Localization and Mapping(SLAM)is a technology for environmental detection and self-positioning of mobile robots in unfamiliar environments.It is the core technology for mobile robots to perform their tasks autonomously,and is also the key technology to enhance the intelligence and flexibility of robots.In addition,SLAM technology is also the basis for the realization of advanced technologies such as unmanned driving and augmented reality.The research and development of SLAM technology can promote the progress of intelligent manufacturing,which has great significance to military defense,manufacturing,service industry,transportation and other fields.Compared with the SLAM algorithm based on particle filter,the SLAM algorithm based on graph optimization has the advantages of smaller cumulative error,higher real-time performance and lower hardware requirements.It is the main research direction of SLAM technology.This paper combines hotspots of the present research.A SLAM algorithm which only needs a single laser sensor has been designed based on the graph optimization theory and the improved point cloud registration algorithm.Then,based on the robot operating system(ROS)as the communication framework,a mobile robot experimental platform and the simulation environment are built by using laser sensor and Mecanum wheel base.The experimental results show that the algorithm achieves a good balance between computational resource requirements and mapping accuracy.In addition,it proves that the built simulation environment has the algorithm verification function,which has certain significance for reducing the development cost of the robot algorithm and shortening the development time.The main contents of this paper include the following aspects:1)The basic solution principle of SLAM problem is introduced.SLAM problem is abstracted into motion model and observation model,which provides mathematical theory basis for SLAM algorithm.The map representation method and robot self-localization method commonly used in SLAM algorithm are summarized.2)The common point cloud registration algorithms are divided into two types: iterative nearest point algorithm and feature-based algorithm.The principles of the two kinds of algorithms are analyzed and their effects are compared.Then an improved point cloud registration algorithm is proposed based on the advantages of these two kinds of algorithms.The experiment proves that the improved point cloud registration algorithm can converge faster under the condition of ensuring accuracy,which provides guarantee for the real-time performance of SLAM algorithm.3)According to the idea of graph optimization,the laser point cloud SLAM algorithm is designed.The improved point cloud registration algorithm is used to estimate the pose transformation in the front end of graph optimization.The data fused with the motion control instructions by unscented Kalman filter to form sub-maps with small cumulative error and more information which saved as grid maps.The back-end of graph optimization takes the poses of sub-maps as the optimization objective.A bag-of-word model is used for loop detection,then the motion model constraints,observation model constraints and loop constraints are added.The optimization equation is solved using the general graph optimization algorithm and the least square method.Finally the poses of sub-maps are adjusted according to the results,then the sub-maps are spliced into global maps.4)The experimental platform of mobile robot is built,and the simulation platform based on real platform is built by using VREP simulation software.The algorithm experiments are carried out on the real platform and the simulation platform respectively.The experimental results confirm the feasibility of the algorithm.It also shows that the simulation platform can replace the real environment to verify the algorithm and reduce the cost of algorithm development.