Research On Localization Method For Autonomous Driving Vehicle

Vehicle positioning technology is a basic issue and has played more and more important role on intelligent driving.A high-precision positioning not only makes it easier for trajectory tracking control of intelligent vehicles,but also achieves vehicle to vehicle(V2V),vehicle to infrastructure(V2I)and vehicle to the city network based on location sharing.It's the only way to achieve intelligent transportation.As proved by DARPA Urban Challenge,it is feasible that vehicles can track its desired trajectory safely and accurately if its positioning accuracy reaches above the order of decimeters with update frequency to be no less than 10 Hz.Although satellite-based positioning method is still the most popular and mature way,but specific to the field of self-driving car,there are major problems in using GNSS(Global Navigation Satellite System)since its signal can often be obstructed by surrounding objects,and its update rate is too low to be adequate.Mounting research has been conducted aimed to improve the performance of GNSS,such as RTK-GPS,inertial navigation system,dead reckoning and visual odometry.The problems remain unresolved,in particular,under “urban canyon” for sufficient long period.Other technologies that utilize radio frequency identification(RFID),ultra wide band(UWB)and 5G are mainly based on wireless communication,which is hardly feasible for practical applications in a large scale.Mounting research has been conducted aimed to improve the performance of GNSS,such as RTKGPS/INS systems,dead reckoning and visual odometry.The problems remain unresolved,in particular,under “urban canyon” for sufficient long period.Other technologies that utilize radio frequency identification(RFID),ultra wide band(UWB)and 5G are mainly based on wireless communication,which is hardly feasible for practical applications in a large scale.Aimed to provide an effective approach for control-oriented applications,this paper proposes a novel method using a high-precision digital map to achieve high-accuracy positioning with fast updating rate.The main research work includes:1)A GPS model with software-centered observation errors is proposed.The error sources of satellite positioning is analyzed,and Gaussian white noise is added to the pseudorange according to the statistical characteristics of the error.As a result,the positioning error is occurred in software and is correlated well with real signals measured by an on-board GPS receiver.2)In order to improve the usability of the satellite positioning system,this paper uses dead reckoning to improve positioning accuracy and update frequency of it.The kinetic model of test vehicle is established and the recursive equation is obtained.The vehicle state parameters in the equation are acquired directly from chassis CAN bus.Aimed to address the nonlinear problem of state equation,UKF(Unscented Kalman Filter)is applied to fuse GNSS and dead reckoning.The experimental results show that the fusion algorithm can effectively improve the positioning accuracy and update frequency with an excellent robustness.3)A high-precision metric-topological map is created.This map can not only overcome the disadvantages of conventional map representation which is single from the form of content and with large data storage,but also benefit for trajectory tracking control,static environment perception,map matching.The topological map uses vehicle trajectory as waypoints which are collected by RTK/INS integrated navigation system.The metric map uses Velodyne HDL-64 E LiDAR sensor to collect point cloud data and is simplified as height difference map.The simplification also satisfies the requirements of map matching and obstacle detection.The two sections of the hybrid map linked by vehicle poses.The cumulative error in process of map building is eliminated by the graph optimization algorithm.4)Based on the high-precision digital map and initial estimation position,the map matching method is proposed to obtain the accurate absolute position of the vehicle.In this paper,the ICP(Iterative Closet Points)algorithm is used to match the point cloud data collected by the current frame with map.Since the estimated position is pre-known and the point cloud data is simplified as height difference map,the accuracy and efficiency of map matching have met the positioning requirements.Theoretic analysis,simulation and experiments show that the proposed method can effectively address the localization problem for autonomous driving vehicle in large-scale urban environment.The high-precision digital map can also be used in trajectory planning and environmental perception.This thesis has some guiding significance and reference value on the application scenario of intelligent vehicle.