Vehicle Model Constraint Based Laser Localization Algorithm For Autonomous Driving In Closed Park

In recent years,due to the increasing maturity of artificial intelligence technology,autonomous driving has become a research hotspot in the automotive field,leading the development wave of the automotive industry.Major automobile manufacturers and Internet companies have followed the trend and conducted technical tests and applications in specific park scenarios,which promotes the commercialization of autonomous driving.The automatic driving system mainly consists of many modules such as environmental perception,positioning and navigation,planning and decision-making,and motion control,among which the positioning and navigation module provides accurate and real-time vehicle pose estimation information for the decision-making and control module,which is an indispensable part of the automatic driving system.At present,most of the automatic driving systems make use of GNSS/INS to achieve high-precision and real-time positioning.However,GNSS may be affected by factors such as occlusion and multipath effects in the park scene,resulting in positioning deviations,which cannot satisfy the accurate positioning requirements of autonomous driving in the park.In recent years,the development of SLAM technology provides a new idea for solving the above problems,and it is expected to reuse the existing sensors of the automatic driving system to realize accurate posit ioning of vehicle under occlusion environment.However,when the existing SLAM technology is applied to the automatic driving positioning system in a closed park,there are still problems such as unstable feature point extraction,easy to fall into a local optimum in pose optimization,and large accumulated drift.In consequence,this subject takes laser positioning system for autonomous driving in closed parks as the research object,adopts dual-neighbor feature extraction strategy to obtain stable corner points and plane points,supplemented by vehicle kinematics model constraints to narrow the pose optimization space,and employ loop detection to reduce the accumulated drift of vehicle pose,aiming to improve the accuracy and smoothness of the laser positioning system for autonomous driving in the closed park scene.The contents of the research mainly include:(1)The architecture design of laser positioning system based on vehicle model constraint.Using steering wheel angle sensor and wheel speed sensor to obtain the speed and angular velocity information of the vehicle,and the vehicle pose change constraint is constructed using the vehicle kinematics model.At the same time,utilize the multi-line lidar to obtain environmental information,and construct the laser odometer residual through data association of point cloud.Based on the residual of the laser odometer and the constraint of vehicle model,the cost function of the positioning system is jointly constructed to achieve tightly coupled pose optimiz ation,and the local curvature information of the point cloud is applied to defined a global descriptor to complete the loop detection and reduce the cumulative drift of vehicle pose.(2)The algorithm design of vehicle kinematics model based on multi-source fusion.The steering wheel angle sensor is used to obtain the steering wheel angle information and calculate the yaw rate while the longitudinal velocity of the vehicle is obtained using the wheel speed sensor.Based on the above-mentioned vehicle data,the vehicle state is predicted through the vehicle kinematics model,and the vehicle kinematics model constraints are constructed in the light of the predicted value.The vehicle pose change constraint provided by the vehicle kinematics model is capable of guiding the gradient direction during the optimization process of gradient descent,and reduce the optimization space,thereby improving the accuracy and smoothness of the laser positioning algorithm,which is conducive to environmental perception and pat h planning.(3)The algorithm design of laser odometer based on double neighborhood curvature feature.The motion distortion of point cloud data is corrected and the filtering operation is performed to obtain the laser point cloud data that is conducive to matching,the local curvature of the point cloud is calculated according to the neighborhood point information,and the dual neighborhood feature extraction strategy is used to obtain stable corner points and plane points.A local map is constructed based on the feature point cloud corresponding to a fixed number of frames,and the residual of the laser odometer is obtained by matching the feature point cloud extracted from the current frame with the local map.The vehicle kinematics model is used to provide a good matching initial value for the ICP algorithm,and the wrong point pairs are eliminated by combining Euclidean distance,local curvature and intensity information of point cloud,which guarantees the accuracy of the matching algori thm.(4)The algorithm design of laser loop detection based on local curvature.The key frame is determined based on the fixed frame interval,the candidate loop frame is obtained on the basis of the distance and time information,the global descriptor of point cloud is constructed by using local curvature,and the similarity between the candidate loop frame and the current frame is calculated according to the discrete Wasserstein metric.After determining the loop frame based on the si milarity,the local map corresponding to the frame is constructed,and the optimized pose of the vehicle is obtained by matching the current frame with the local map.Time consistency and spatial consistency are used to verify the accuracy of the loop,calculates the cumulative drift of the vehicle pose and uses the interpolation algorithm to update the historical pose,thereby effectively reducing the cumulative error of vehicle pose.(5)Real vehicle verification based on closed park data.Collect multiple sets of park data through the autonomous driving platform car and Robot Operating System software platform,and contrastive analysis is performed on the test results to fully verify the effectiveness of the laser positioning algorithm.The experimental results show that compared with the current prevalent laser localization algorithm ALOAM,the laser localization system based on vehicle kinematic model constraint is able to provide more accurate and smooth vehicle positioning information.