A Multimodal Approach for Determination of Vehicle Position

Transcription

1 A Multimodal Approach for Determination of Vehicle Position Artis Mednis Digital Signal Processing Laboratory, Institute of Electronics and Computer Science, 14 Dzerbenes Str., Riga, LV 1006, Latvia, Abstract. One of the most important tasks during development of hardware/software systems for assisted and automatic driving is determination of vehicle position with sufficient accuracy in real time. Most of experimental systems developed yet are based on relatively expensive Real Time Kinematic technique which primary usage is land and hydrographical surveys with centimetre level accuracy. The paper is describing technique for determination of vehicle position, based on multimodal data sources with subsequent data integration. Data acquisition is performed using several GNSS receivers with SBAS capability as well as stationary reference stations. Data acquisition and integration methods are discussed as well as their performance analyzed using data from real world experiments. Key words: vehicle position, multimodal, data integration, GNSS, SBAS, RTK over IP 1 Introduction Vehicle driving process is one of the human s activities characterized by large amount of individual decisions. Decisions in this case should be taken especially carefully because they affect not only own vehicle and passengers safety but also safety of surrounding vehicles and humans. Due evolution of various transportation types the optimization of traffic flow becomes more and more important. This task can be solved in both ways - due more advanced traffic flow management as well as more advanced dissemination of traffic information for each individual driver and/or vehicle. Solutions developed for solving of this task include also appropriate hardware/software systems. These systems could be divided in two main groups - systems for assisted driving and systems for automatic driving. Systems from first group perform a part of activities of driving process and give already pre-processed information to driver This is an author-created version. The final publication is available at /

2 and/or vehicle. Systems from second group act independently and perform vehicle driving process without human interaction. Depending on their architecture and functionality systems for assisted or automatic driving are intended to perform several specific tasks. One of them is determination of vehicle position with accuracy acceptable for successful operation of whole system. In the case of standalone automatic driving this information is used as input data for motion planning but in case of cooperative assisted driving this information is shared among several vehicles. Common approach for determination of position of an outdoor object is usage of generic GNSS ([1], p.2) receiver. Unfortunately accuracy of this approach is affected by several factors such as buildings in the urban area, trees in the forest, weather etc. Therefore this common approach is suitable for navigation of a person or a vehicle in the case when a human uses acquired data but is not suitable for assisted or automatic driving when a computer uses acquired data. This paper is describing technique for determination of vehicle position with accuracy, suitable for particular assisted driving solution. This technique is based on data acquisition using several GNSS receivers with SBAS ([2], p.157) capability. Correction data from stationary reference stations received through wireless Internet connection is used as additional input. Related work is described in Section 2. Approach is discussed in Section 3. The evaluation of proposed approach includes a series of real world experiments, analyzed in Section 4. The final section presents the conclusion that proposed approach is suitable for assisted driving system developed for GCDC competition vehicle. 2 Related Work Systems which primary functionality is determination of exact vehicle position are components of almost all more advanced systems developed for automatic or assisted driving competitions. Typical examples of automatic driving competitions are events organized by Defense Advanced Research Projects Agency (DARPA) such as Grand Challenges in 2004 and 2005 [3] as well as Urban Challenge in 2007 [4, 5]. A typical example of assisted driving competition is Grand Cooperative Driving Challenge (GCDC) in 2011 [6]. Different competition types are characterized by different scenarios and therefore different requirements for system parameters including accuracy of vehicle position. Two most successful position determination approaches from before mentioned competitions are described in this Section. Team Tartan Racing [7], which represents Carnegie Mellon University in cooperation with General Motors, Caterpillar and Continental AG, won DARPA Urban Challenge. Their vehicle position determination solution was based on Applanix POS-LV [8] device consisting of GNSS receiver and inertial measurement unit. Additional sensors such as lidars, radars and cameras were used to ensure acceptable level of safety in mocked urban area.

3 Team AnnieWAY [9], which represents Karlsruhe Institute of Technology, won Grand Cooperative Driving Challenge. Their vehicle position determination solution, which was similar to previously mentioned, was based on OXTS RT 3003 [10] device consisting of GNSS receiver and inertial measurement unit. Additional sensors such as high definition laser scanner and multiple cameras were used to ensure safe distance to vehicle driving in front. 3 Approach There exist several sources of data errors that affect position determination using GNSS signals. These sources include signal propagation in Earth s ionosphere and troposphere, reflected multipath signals, parameters of satellite orbits as well as clock drift both on transmitters and receivers side. All mentioned error sources could be divided in two main groups - local sources and global sources. Sources from first group are characterized by similar influence on the determined position in restricted geographical area. Sources from second group are characterized by similar influence on the determined position in wide geographical area ([11], p ). A simple approach to minimize data errors from local sources is making of several consecutive position measurements and subsequent calculation of average position values. This method is suitable only if calculation of the position is performed for stationary object therefore not suitable for calculation of the position for vehicle in motion. Proposed approach assumes quasi-simultaneous position measurements using several generic SBAS receivers instead of several consecutive measurements using one generic SBAS receiver. Calculation of average position data is performed immediate after data acquisition. A simple approach to minimize data errors from global sources is usage of stationary placed dedicated local reference station in known position nearby as well as advanced multi-frequency (L1/L2) receiver. This method assumes usage of specific and therefore relatively expensive RTK ([12], p.15) hardware. The need for deployment of such specific and expensive hardware in wide area makes this method not applicable for real-world assisted and automatic driving experiments. Proposed approach assumes usage of public available RTK base stations and receiving of correction data using wireless Internet connection (Fig. 1). 4 Evaluation After studies of literature about several types of vehicle positioning systems it was decided to build GCDC competition vehicle positioning system using multimodal approach. In this case data for position calculation should be acquired from several SBAS receivers as well as from inertial measurement unit (IMU). Such approach looked potentially attractive due relative low costs and in the

4 Fig. 1. Architecture of proposed vehicle position determination approach. First step includes data acquisition from several SBAS capable GNSS receivers and their averaging, second step includes position correction using data from public reference stations received over wireless Internet connection. same time accuracy sufficient in the context of published competition specification [13]. Selected SBAS receivers were tested using real world data in several modes including static data acquisition mode, dynamic data acquisition mode as well as correction data usage mode. First experiment was carried out to estimate typical position deviation of single SBAS receiver in static data acquisition mode. During this experiment Magellan explorist XL receiver [14] was placed in fixed position on the roof of a stationary parked vehicle. Session of position data acquiring was 1 hour long and position data was recorded 1x per second. During this session SBAS receiver was in the DGPS mode 100% of all time and every position was calculated using data from at least 9 satellites. Statistical analysis of recorded position data showed that maximal receiver position deviation in the west-east direction (lon) is 1.51 m and in the north-south direction (lat) 1.81 m (Fig. 2 - on the left). According to 2DRMS corresponding range of 98.2% ([15], p.153) receiver position deviation in the west-east direction (lon) is 1.51 m and in the north-south direction (lat) 1.48 m. Graphical analysis of recorded position coordinates and their distribution through deviation area showed that there are at least two main areas with high recorded positions concentration located in opposite deviation area corners (Fig. 2 - on the right). Such distribution does not correspond with Gaussian distribution and allows making of an assumption about influence of global data error sources what could be mitigated using additional correction data. Next experiment was carried out to estimate typical position deviation of two SBAS receiver system in static data acquisition mode. There was an assumption about potential decreasing of system position deviation area due usage of two independent measurement devices and combination of simultaneously acquired position data. Setup of this experiment was similar to previous performed single SBAS receiver test except an additional SBAS receiver Magellan explorist 210

5 Fig. 2. Static data acquisition mode test using single SBAS receiver. On the left - position deviation track, on the right - position deviation distribution. [16] placed on the roof of the same stationary parked vehicle but in opposite corner. During this session both SBAS receivers were in the DGPS mode 100% (XL) and 99.97% (210) of all time and every position was calculated using data from at least 7 (XL) and 8 (210) satellites. Statistical analysis of recorded position data showed that maximal receiver position deviation in the west-east direction (lon) is 1.31 m (XL) and 2.31 m (210) and in the north-south direction (lat) 1.67 m (both XL and 210) (Fig. 3 - on the left). According to 2DRMS corresponding range of 98.2% receiver position deviation in the west-east direction (lon) is 1.21 m (XL) and 1.91 m (210) and in the north-south direction (lat) 1.67 m (XL) and 1.48 m (210). Graphical analysis of recorded position coordinates and their distribution through deviation area showed that there is one main area with high recorded positions concentration for each receiver used but these areas are located in opposite deviation area corners (Fig. 3 - on the right). Combination of simultaneously acquired position data pairs was performed, calculating average latitude and longitude values. Statistical analysis of calculated position data showed that maximal system position deviation in the westeast direction (lon) is 1.31 m and in the north-south direction (lat) 0.74 m (Fig. 4 - on the left). According to 2DRMS corresponding range of 98.2% system position deviation in the west-east direction (lon) is 1.11 m and in the north-south direction (lat) 0.55 m. Graphical analysis of calculated position coordinates and their distribution through deviation area showed that there is one main area with high recorded positions concentration what corresponds with Gaussian distribution. Decreasing of the position deviation area allows making of a conclusion about improvement of position determination accuracy against usage of single SBAS receiver. Next experiment was carried out to estimate typical position deviation of two SBAS receiver system in dynamic data acquisition mode as well as perform data capture from inertial measurement unit. Before this experiment two Magellan

6 Fig. 3. Static data acquisition mode test using two SBAS receivers. On the left - position deviation tracks, on the right - position deviation distributions. At the top - data from Magellan explorist XL, at the bottom - data from Magellan explorist 210. Fig. 4. Static data acquisition mode test using two SBAS receiver system. On the left - position deviation track, on the right - position deviation distribution.

7 explorist receivers - XL and 210 were placed in fixed position on the dashboard of the vehicle. During this experiment vehicle carried out 67 km long route within 1 hour. Route selected for this experiment consisted of 58 km of highway with maximal permitted speed 90 km/h as well as 9 km of city streets with maximal permitted speed 50 km/h (on several fragments - 70 km/h) (Fig. 5). Maximal real speed of vehicle during this experiment was 63 km/h on a city street as well as 103 km/h on the highway. Position data was recorded 1x per second but data from inertial measurement unit (3D accelerometer and 2D gyroscope) 10x per second. During this session both SBAS receivers were in the DGPS mode 54.47% (XL) and 88.57% (210) of all time and every position was calculated using data from at least 6 (XL) and 5 (210) satellites. Comparison of DGPS mode time and available satellites count with according parameters from static data acquisition mode experiments lets make a conclusion about more severe data acquisition conditions. Fig km long route selected for dynamic data acquisition mode experiment. First combination of the position data from both SBAS receivers was performed using granularity identical to the data acquisition rate - 1 second. Statistical analysis of calculated position data showed that maximal system position deviation in the west-east direction (lon) is 9.79 m and in the north-south direction (lat) 6.50 m (Fig. 6). According to 2DRMS corresponding range of 98.2% system position deviation in the west-east direction (lon) is 9.08 m and in the north-south direction (lat) 5.01 m. Average deviation distances in the west-east direction (lon) is 4.71 m and in the north-south direction (lat) 2.16 m. Distribution function of deviation distances is shown in Fig. 7. Position deviation during dynamic data acquisition mode is affected not only by receivers position calculation errors but also by speed of the moving vehicle. In this case vehicle speed during test could be responsible for position deviation up to 25 meters. Real position deviation values were less then 10 meters. GCDC competition specification includes the requirement that prescribes vehicle position data actualization 10x per second. To comply with this require-

8 Fig. 6. Dynamic data acquisition mode test using two SBAS receiver system. System position deviations using data combination with granularity 1 second. Fig. 7. Dynamic data acquisition mode test using two SBAS receiver system. Distribution functions of system position deviations using data combination with granularity 1 second - on the left west-east direction (lon), on the right north-south direction (lat). ment other data combination approach using granularity of 1/10 seconds range was performed. To reach this granularity, position data from one receiver was used unaltered, but position data from other receiver was transformed with the aim to obtain vehicle position data for the same moment of time (Fig. 8). Statistical analysis of calculated position data using granularity 1/10 second showed that maximal system position deviation in the west-east direction (lon) is 2.42 m and in the north-south direction (lat) 3.71 m (Fig. 9). According to 2DRMS corresponding range of 98.2% system position deviation in the westeast direction (lon) is 1.90 m and in the north-south direction (lat) 2.86 m. Average deviation distances in the west-east direction (lon) is 0.90 m and in the north-south direction (lat) 1.26 m. Distribution function of deviation distances is shown in Fig. 10. Maximum increasing of the position deviation due moving

9 Fig. 8. Vehicle position data combination with the granularity of 1/10 seconds range. Positions from SBAS receiver B recalculated according timestamps from SBAS receiver A. vehicle speed in this case could be up to 2.5 meters. 94% of real measurements were under this value. Fig. 9. Dynamic data acquisition mode test using two SBAS receiver system. System position deviations using data combination with granularity 1/10 second. Additional improvements of two-receiver system performance in dynamic data acquisition mode could be achieved after next step that includes usage of acceleration data acquired using inertial measurement unit. During GCDC Technology Workshop ( , Helmond, Netherlands), author s colleague got an information about specific methodology - RTK over IP, potentially useful for positioning accuracy improvement. This methodology is based on usage of additional correction data received in real time through wireless Internet connection. To verify this methodology as potential improvement of developed GCDC competition vehicle equipment, several base stations

10 Fig. 10. Dynamic data acquisition mode test using two SBAS receiver system. Distribution functions of system position deviations using data combination with granularity 1/10 second - on the left west-east direction (lon), on the right north-south direction (lat). such as TORA0 (Tartu, Estonia) and TITZ1 (Titz, Germany) for real world experiments were identified. To ensure access to resources of these base stations accounts in corresponding data distribution networks - [17] and [18] were created. Next experiment was carried out to estimate how the usage of correction data from one or several RTK base stations improves position accuracy if the local position measurements are performed using generic single-frequency (L1) receiver instead of a dedicated multi-frequency (L1/L2) receiver. Local position measurements in Cesis were performed using Magellan XL SBAS receiver. Correction data were obtained from two LatPos system [19] base stations located in Sigulda and Valmiera. Both of them were about 30 km from stationary placed SBAS receiver (Fig. 11). First, combination of locally acquired position data and correction data from each one RTK base station was performed. Statistical analysis of corrected position data showed that maximal corrected position deviation in the west-east direction (lon) is in the range m and in the north-south direction (lat) in the range m. According to 2DRMS corresponding range of 98.2% corrected position deviation in the west-east direction (lon) is in the range m and in the north-south direction (lat) in the range m. Graphical analysis of corrected position coordinates and their distribution through deviation area (Fig. 12) showed that there is a tendency towards one main area with high recorded positions concentration in the middle of deviation area what could be appreciated as improvement against distribution in the case of uncorrected position data (Fig. 2 - on the right). Next, combination of locally acquired position data and correction data from both RTK base stations was performed. Statistical analysis of corrected position

11 Fig. 11. Correction data from RTK base stations. On the left - data from station in Sigulda, on the right - data from station in Valmiera. In the middle - local position measurements from stationary placed SBAS receiver in Cesis. Fig. 12. Corrected position deviation distribution. On the left - correction data from RTK station in Valmiera used, on the right - correction data from RTK station in Sigulda used. data showed that maximal corrected position deviation in the west-east direction (lon) is 1.46 m and in the north-south direction (lat) 1.52 m. According to 2DRMS corresponding range of 98.2% corrected position deviation in the west-east direction (lon) is 1.31 m and in the north-south direction (lat) 1.30 m. Graphical analysis of corrected position coordinates and their distribution through deviation area (Fig. 13) showed that there is a strong tendency towards one main area with high recorded positions concentration in the middle of deviation area. Decreasing of the position deviation area allows making of a conclusion about improvement of position determination accuracy against usage of single SBAS receiver without external correction data.

12 Fig. 13. Corrected position deviation distribution using correction data from two RTK stations located in Valmiera and Sigulda. 5 Conclusion and Future Work Author proposed vehicle position determination approach that includes usage of several SBAS receivers as well as data from local reference stations. This approach was evaluated on particular application - developing GCDC competition vehicle positioning system. Vehicle position determination was performed using multimodal data sources and combination of their data. Author performed several real world experiments using selected position data acquisition devices. The experimental results were evaluated by statistical and graphical analysis of position deviation distribution area. The results show, that proposed vehicle position determination approach decreases range of position deviation in the north-south direction (lat) from 1.48 m to 0.55 m 2DRMS due usage of two SBAS receivers as well as from 1.48 m to 1.30 m 2DRMS due usage of correction data from RTK base stations. Changes of position deviation in the west-east direction (lon) is less notable - from 1.21 m to 1.11 m 2DRMS due usage of two SBAS receivers as well as from 1.51 m to 1.31 m 2DRMS due usage of correction data from RTK base stations. Therefore proposed approach is potentially suitable for GCDC competition vehicle [20] which position accuracy should be 1 meter 2DRMS or better. The future work includes experiments with increased number of SBAS receivers and evaluation of their influence on position determination accuracy as well as improvement of position calculation process using data from the inertial measurement unit Acknowledgments. This work was supported by European Social Fund grant Nr. 2009/0219/1DP/ /APIA/VIAA/020 R&D Center for Smart Sensors and Networked Embedded Systems. Author thanks to Reinholds Zviedris and Ojars Krumins for SBAS receivers provided for testing purposes.

13 References 1. Gleason, S., Gebre-Egziabher, D.: GNSS Applications and Methods. GNSS Technology and Applications. Artech House (2009) 2. Samama, N.: Global Positioning: Technologies and Performance. Wiley Survival Guides in Engineering and Science. Wiley-Interscience (2008) 3. Buehler, M., Iagnemma, K., Singh, S.: The 2005 DARPA Grand Challenge: The Great Robot Race. Springer Tracts in Advanced Robotics. Springer (2007) 4. Defense Advanced Research Projects Agency: DARPA Urban Challenge. http: //archive.darpa.mil/grandchallenge/index.asp 5. Buehler, M., Iagnemma, K., Singh, S.: The DARPA Urban Challenge: Autonomous Vehicles in City Traffic. Springer Tracts in Advanced Robotics. Springer (2010) 6. GCDC Organization: Grand Cooperative Driving Challenge. net/ 7. Urmson, C., Anhalt, J., et al.: Tartan Racing: A Multi-Modal Approach to the DARPA Urban Challenge. TechPapers/Tartan Racing.pdf 8. Applanix Corp.: POS LV Specifications. (January 2011) 9. Geiger, A., Moosmann, F., et al.: Team AnnieWAY s Entry to GCDC current/gcdc Final% 3A teams videos and final papers/annieway 10. Oxford Technical Solutions: RT Inertial and GPS Measurement Systems. User Manual. (December 2010) 11. Grewal, M.S., Weill, L.R., Andrews, A.P.: Global Positioning Systems, Inertial Navigation, and Integration. Wiley-Interscience (2007) 12. LaMarca, A., Lara, E.: Location Systems: An Introduction to the Technology Behind Location Awareness. Synthesis Lectures on Mobile and Pervasive Computing. Morgan & Claypool Publishers (2008) 13. GCDC Organization: GCDC 2011 Rules & Technology Document Final Version. (April 2011) 14. Thales Navigation, Inc.: Magellan explorist XL Reference Manual. (September 2005) 15. Farrell, J.: Aided Navigation: GPS with High Rate Sensors. McGraw-Hill Professional Engineering: Electronic Engineering. McGraw-Hill (2008) 16. Thales Navigation, Inc.: Magellan explorist 210 Reference Manual. (June 2005) 17. Federal Agency for Cartography and Geodesy: EUREF-IP Ntrip Broadcaster Federal Agency for Cartography and Geodesy: IGS-IP Ntrip Broadcaster. http: // 19. Latvian Geospatial Information Agency: LatPos. index.php?lang=2&cpath=2&txt id= Strazdins, G., Gordjusins, A., et al.: Team Latvia GCDC 2011 Technical Paper. - current/gcdc Final% 3A teams videos and final papers/latvia