TAXISAT: A Driverless GNSS Based Taxi Application Capable of Operating Cost Effectively

O. Otaegui, O. Desenfans, V. Barreau, L. Plault, A. Lago

Abstract:	Although automatic and driverless vehicles do operate nowadays most of these systems are currently guided by railways, fiberglass or magnetic wires. More recently, a few demonstrations have been undertaken in order to demonstrate the capabilities of satellite based navigation systems to clear the roads from such heavy infrastructures. TAXISAT aims at demonstrating a low cost, reliable and secure driverless vehicle hybridizing EDAS, GNSS and visual SLAM technologies improving the integrity, reliability and availability of the unmanned driving services even in case of temporary canyoning or tunnelling. The main objective of TAXISAT is to develop a low cost localization and navigation solution for driverless vehicles dedicated to the transport of passenger in open sites In order to enable a safe and comfortable journey to passengers, the vehicle shall be capable of following accurately and smoothly its trajectory. That can only be achieved if the positioning-guidance chain is of high quality. Reliability and integrity of the position are at least as important as its accuracy and precision. Previous experience demonstrates that reliability and integrity are fundamental for a commercial application. Indeed, as soon as the GPS signal is lost, IMU data are not corrected by D-GPS (accuracy of about < 2 cm) anymore, and the positioning system starts drifting. In order to prevent safety issue, the system automatically stops (allowing only manually control). While a couple of identified reasons explain GPS signal losses (such as growing tree, or masked constellation), it remains inexplicable signal losses. Solutions can be found in the first case. Unfortunately, in the latter case there is no solution but waiting. In this sense TAXISAt will offer a new solution to be operative in all-type of conditions. Requirements from potential customers state that the positioning performances shall be such that vehicles strictly remain in their circulation lanes (positioning error lower than about 2m), and that vehicle shall be able to operate at least 95% of the time (reliability objective). TAXISAT solution is composed of two subsystems: a positioning module (that provides the vehicle position) and a navigation module (that hosts the vehicle control laws so that the vehicle follows the defined trajectory). One of the main contributions of TAXISAT is the use of video based positioning systems. The input of the video positioning module aims at enhancing its performance level compared to usual INS/GNSS hybrid system. The accuracy as well as the reliability, the robustness and the continuity of the TAXISAT hybrid-positioning module is increased thanks to the use of video analysis. Video based positioning constitutes an additional source of data which, in itself, is already an improvement of the hybridisation process performances. In addition, thanks to an expected precision of about 10 cm and a rate of about 25 Hz, video will increase the efficiency of the INSs drift adjustment and correction which will increase the system accuracy. TAXISAT video based positioning provides information about transversal displacements and estimations of velocity and orientation from a low-cost stereo pair camera system, and the associated video-based analysis algorithms. Novel trends on video positioning like Simultaneous Localization and mapping (SLAM) have been included in order to obtain more robust and reliable positioning information through the application of inference probabilistic methods. Visual SLAM techniques implies the simultaneous computation of the pose of the cameras and therefore of the vehicle itself and the structure of the scene. Nevertheless, there are many options regarding the type of reconstruction of the scenario. Typically, only the 3D position of a sparse set of point-wise key points or landmarks are obtained. Therefore, it is not always possible to determine semantic knowledge of the scenario from these sets. Special efforts have been made to obtain dense landmarks reconstruction such that texture mapping can be applied from images to 3D meshes. This information also is valuable for hybridising with GIS information and computing and accurate map of the scene that enhances the accuracy when computing the trajectory of the vehicle. To meet the aforementioned requirements, TAXISAT positioning subsystem implements an extended Kalman filter--based on tight hybridization algorithm. Different level of hybridization with GNSS could have been implemented; loose, tight, ultra-tight, and deep according to GNSS data used, respectively the PVT fix (Position – Velocity – Time), the raw data (pseudoranges and Doppler), and the correlators measurements. Most of the existing solutions rely on loose hybridization, which requires the PVT fix. The calculation of the PVT implies that at least 4 satellites be in view. In the TAXISAT context, where urban area is of major interest, such approach is not satisfying. Indeed, due to canyoning or tunneling, it is not unusual that the number of satellite in view drops below this minimum, which leads to the loss of the PVT, and, as a consequence, the loss of capability to adjust other sensors drifts. In order to assess performances prior tests on the field, complete subsystems simulation have been performed. Both nominal and non-nominal (i.e., with perturbations such as multipath) situation have been analyzed. The accuracy and the robustness sensitivity to latency on measurement time stamping and sensor quality have been investigated. To do so, a Monte-Carlo approach has been followed to enable statistical analysis of results over large sets of data. Then threes test and demonstration sessions have been planned: April2013 in San Sebastian technology Park, Spain, in July2013 at the Spa-Francorchamps’ race circuit, Belgium, and at Vulcania attraction park, France. It is to be noted that Vulcania is already equipped with three automated guided vehicles, which face reliability issue due to lost of GNSS based IMU’s drift adjustment. Simulation results are very promising and demonstrate that the performance requirements could be met with the proposed lowcost sensors based solution. This project is partially funded by the EU 7th Framework Programme.
Published in:	Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013) September 16 - 20, 2013 Nashville Convention Center, Nashville, Tennessee Nashville, TN
Pages:	1634 - 1641
Cite this article:	Otaegui, O., Desenfans, O., Barreau, V., Plault, L., Lago, A., "TAXISAT: A Driverless GNSS Based Taxi Application Capable of Operating Cost Effectively," Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013), Nashville, TN, September 2013, pp. 1634-1641.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In