Abstract—Autonomous driving requires continuous and reliable centimeter level positioning accuracy for acceptable lane level navigational performance. Centimeter level positioning accuracy cannot be achieved using a conventional DGPS/DR. To deal with the above problem this paper proposes the novel and effective bias/shift estimation of DGPS/DR. On this paper, a surveyed precise map is used that is consisting of waypoints with centimeter level positioning accuracy for the lateral and longitudinal directions. The time history of DGPS/DR waypoints are compared with the closest set of waypoints from surveyed precise map. Both a straight line fitting method and a sliding curve method have been used in order to match the shape of the DGPS/DR trajectory with surveyed precise map. For lateral and longitudinal DGPS/DR bias estimation, we have adopted a disturbance observer. Finally, experiments were conducted to prove the feasibility of the proposed algorithm for the shift estimation of DGPS/DR. This paper also compares the experimental results of GPS/DR with the ones using RTK GPS/DR during autonomous driving. Index Terms—RTKt, DGPS bias, Map based DGPS bias estimation, DGPS based autonomous driving. I. INTRODUCTION In the recent time promising demonstrations of self- driving car in real conditions have been carried out using market available sensors [1]. For most of the self-driving car demonstrations a custom made or informative digital map of the route to be followed by professional map making companies has been used. For implementing the reliable and continuous navigation capability, a vehicle satellite positioning system is augmented with other sensors such inertial measurement unit, odometer and exteroceptive sensors such as camera and LIDAR [2], [3]. Differential correction obtained by using GPS pseudo-range code (DGPS) can achieve an accuracy within 1-5m, on the other hand, the usage of GPS carrier phase information such as real time kinematics (RTK) GPS can achieve centimeter level positioning accuracy. In case of loss of GPS signal due to environmental causes leads to the degradation of positioning accuracy [4], [5]. In such scenario the bias or the shift in the position should be estimated online and compensated for safety reasons. Map matching algorithms have been used extensively in prior works to match the position from the positioning system to one on the road network in the map. Usually, the time history of the trajectory is matched with the shape of Manuscript received July 6, 2017; revised August 20, 2017. S. S. Rathour, Ali Boyali, Lyu Zheming, Seiichi Mita, Vijay John are with Research Center for Smart Vehicle, Toyota Technological Institute, Nagoya, Japan (e-mail: {swarn,ali, smita, vijayjohn}@toyota-ti.ac.jp, lvzheming@gmail.com). the road network. Map matching technique generally include geometrical analysis such as point to point matching, point to arc matching and curve to curve matching, probability theory, extended Kalman filter, and hidden Markov model [6], [7]. In order to reduce localization error, [8]-[10] proposed the method that uses lane detection method to improve the GPS/DR error estimation. However, in case of no lane visual information such method is susceptible of noise addition. Various authors [4], [5] proposed the use of enhanced map-matching algorithm for urban environments, using topological information of digital map, as well as historical information of the route followed by a fuzzy rule for real time application. Autonomous driving requires continuous and reliable centimeter level positioning accuracy for acceptable lane level navigational performance. Map matching algorithm augmented with exteroceptive sensors such as camera and LIDAR can be used to derive DGPS/DR lateral bias, but longitudinal bias is difficult to estimate [1]. In order to overcome the limitation of map matching algorithm, as well as dependency of lane information and visible feature for exteroceptive sensors for bias estimation we propose the use of surveyed precise map built by using RTK GPS with centimeter level positioning accuracy. The surveyed precise map consists of waypoints at 10 cm distance interval. In order to predict the DGPS/DR lateral and longitudinal bias we use the time history of position trajectory. This paper has been divided into five sections. The first section briefly introduces the objective of the paper followed by the brief literature survey and a brief introduction of the paper. Second section defines the terminology used and, mathematical formulation of disturbance observer for bias estimation. Third section deals with the formulation of straight line fit algorithm and curve to curve matching. Fourth section explains the experimental results. Finally, last section is for conclusion. II. MAP-MATCHING ALGORITHM A. Problem Definition Two coordinated system, earth fixed coordinate system  and body fixed coordinated system  has been used (Fig. 1) for defining the variables and control application. For map matching three definitions for set of waypoints representing section of the map  (Fig. 1), set of waypoints representing the vehicle trajectory  and finally full map  are defined as follows: Test Segment: Test segment,  =  1 , 1 = 1,2 … is the time history of the position trajectory from DGPS/DR as shown in Fig. 1. A Map-based Lateral and Longitudinal DGPS/DR Bias Estimation Method for Autonomous Driving S. S. Rathour, Ali Boyali, Lyu Zheming, Seiichi Mita, and Vijay John 67 International Journal of Machine Learning and Computing, Vol. 7, No. 4, August 2017 doi: 10.18178/ijmlc.2017.7.4.622