Research On 3D Construction Of Wheeled Robot Based On Graph Optimization

Constructing 3D map is the primary problem of 3D navigation and obstacle avoidance perception,and it is the premise to realize robot autonomy and intelligence.For the autonomous behavior of wheeled robots,LIDAR sensors have always been one of the preferred sensors for building maps.Their excellent performance and frequency can ensure that the construction can obtain relatively reliable data input.However,in order to construct a high-precision map,it is necessary to use high-precision pose information to provide optimization for it,especially in urban scenes,because the complexity of the urban scene causes the GPS(Global Position System)signal to be blocked by the building.Data reliability is degraded or even completely unreliable.Aiming at this problem,this paper completed the three-dimensional map construction of wheeled robot through SLAM(Simultaneous Location and Mapping)method,and optimized the construction result to ensure the high precision of 3D construction.The main contents of the thesis are:1.Aiming at the drift problem of complex scene construction and the low accuracy of wheeled robot pose,the three-dimensional map construction technology based on SLAM is studied.By abstracting the robot pose into a graph form,the constraint conditions are constructed by the probability distribution of the wheel robot pose,and the data association is formed.Then,the wheel robot trajectory is calculated based on the map mapping algorithm to complete the map work.2.Aiming at the problem of continuous accumulation of map errors during the construction process,a three-dimensional mapping optimization method based on frame to sub-map is studied.In a large-scale scenario,the frame-to-sub-map method is used to optimize the mapping results.The sub-map and the scan frame are matched by the nonlinear optimization method.An improved branch and bound algorithm is proposed,and the algorithm is used to determine the closed-loop.To a certain extent,the efficiency and accuracy of the search constraints are improved,and the optimization of the three-dimensional map construction is realized.3.Aiming at the problem of low precision of sensor data in SLAM technology,a three-dimensional mapping optimization method for correcting sensor information is studied.The method is divided into two steps.The first step is the high-frequency processing of the wheeled robot motion estimation process,and the second step is the low frequency(one order of magnitude lower than the high frequency)to perform map matching for correcting the sensor information,which guarantees the basis of the calculation efficiency.The construction error is reduced,and the accuracy and efficiency of the three-dimensional construction are improved to some extent.4.Based on the above research,the method of frame-to-sub-map and correction of sensor information is carried out by a wheeled robot platform equipped with laser,radar,IMU(Inertial Measurement Unit)and other sensors.Extensive experiments have been carried out in the outdoor environment to verify the accuracy and effectiveness of the proposed method.The results show that the method achieves two-dimensional and three-dimensional mapping of outdoor scenes with a resolution of 0.05 m and an average error rate of about 0.5% in a regular building scene.It has good feasibility and accuracy for autonomous driving and other labor.Intelligent field-related 3D positioning,3D perception,etc.provide a reliable source of data.