Research On Graph-based SLAM Algorithms For Mobile Robots

With the continuous iteration of science and technology,trends of facility clustering and equipment automation are increasingly obvious in numerous fields,such as manufacturing,transportation and advanced agriculture,which has prompted the rapid development of robot industry as never before.Due to the reasonable cost and convenient driving,wheeled mobile robots have played a significant role in many fields.In order to achieve autonomous localization and navigation in any environment of robots,Simultaneous Localization and Mapping(SLAM)is created.For the possible use scenarios of mobile robots,the thesis analyzed the robot's chassis structure that can wildly adapt extensive environments,and the major hardware of the chassis was briefly introduced.On the basis of the chassis mechanism modeling,the kinematics model of the four-wheeled omnidirectional mobile robot was constructed,and the robot odometry was obtained from the dead reckoning.Aimed at the non-structural factors that may exist in the unknown operating environment of the robot,the graph-based two-dimensional SLAM algorithm was proposed for the factor graph construction by sensor observation data,and the nonlinearly global optimization was carried out in the back-end.Since the observation data is a kind of time series,the thesis proposed the method of using local incremental QR matrix decomposition to achieve Incremental Smoothing and Mapping(iSAM).The incremental update method employed by iSAM ensured the real-time operation on the existing hardware and the reliability of the solution result was guaranteed through the interval global optimization on the back-end.The front-end matching and closed-loop detection was achieved by detecting the similarity of laser scanning frames.In addition,insert of the closed-loop factor detected by similarity enabled the closed-loop incorporate into the global optimization problems under the graph-based optimization structure.Based on the Virtual Robot Experimentation Platform(V-REP),the existing two-wheeled differentially driven mobile robot and the utilization of the Solidworks plug-in,the thesis built a joint experimental environment for the communication with Robot Operation System(ROS).Using the proposed algorithm and the mainstream open source algorithm,the thesis constructed the online map on the virtual simulation environment and the dataset respectively,and compared the environment maps constructed by different algorithms;the main indicators of the map through third-party software were also qualitatively evaluated.The good robustness of the proposed algorithm on different data was proved.In addition,in the outdoor non-structural environment,the online environment construction test was carried out using the omnidirectional mobile chassis,and the capability of the proposed algorithm to construct the map in a dynamic and unstructured large environment was also verified.