Map-based Localization Of Self-Driving Car Using Multi-sensor In Diverse City Scenes

With the development of robotics technology,the progress of social productivity has been continuously promoted.As a special ground mobile robot,low-speed unmanned vehicle has received extensive attention in recent years.At present,most of the localization researches for mobile robots are based on simple limited scenarios or using expensive sensors to solve complex and diverse environmental problems,which greatly limits the large-scale commercialization of low-speed unmanned ground vehicles(UGV).In this paper,a low-cost and high-precision UGV positioning scheme is proposed,which can make UGV achieve centimeter-level positioning accuracy in diverse environments.In this scheme,three sensors,low-beam three-dimensional lidar,low-cost inertial sensor and global positioning system,are used to estimate the position and attitude of unmanned vehicle.The point cloud scanned by lidar is registered with the existing global environment feature map by normal distribution transformation,and the global position of the radar can be obtained.The inertial sensor is used to measure the change of the motion state of the unmanned vehicle globally.The positioning system is used to provide global constraints for pose estimation of unmanned vehicle.Finally,an unscented Kalman filter is used to fuse the measured data from the three sensors to obtain an accurate and stable estimation of the state of the UGV positioning system.In software implementation,this paper uses the communication mechanism of ROS to design a multi-node distributed software architecture,which includes the sensor device driver layer and application layer according to the function.The device driver layer transfers the real-time data to the upper application program by running the driver of each sensor in parallel.The application layer completes the positioning work through the cooperation of the map server and the position and attitude estimator.Considering the time disorder of multi-sensor data receiving,a software framework is proposed to realize time alignment of multi-sensor data.In addition,this paper also proposes a method to analyze the positioning accuracy through the stability of the positioning results in the case of unstructured outdoor and lack of true ground position.Based on practical experience,this paper also makes some engineering optimization for the current scheme,using simplified motion model to reduce the interference of sensor measurement noise to the system,and using map dynamic preloading technology to reduce the resource consumption of system calculation.The scheme proposed in this paper has been tested many times in a variety of environments,and has been deployed in a low-speed unmanned vehicle.The experimental results show that the positioning system can run stably on the embedded platform,and the system can achieve the required centimeter-level positioning accuracy,thus providing a guarantee for the large-scale production of low-speed unmanned vehicles.