Session A1, Paper #1
RedBlade: Miami University´s Multi-Functional Autonomous Robot
R. Wolfarth, S. Taylor, A. Wibowo, B. Williams, V. Gatto, J. Morton, P. Jamieson, Miami University
Autonomous vehicles capable of performing many functions with accuracy and reliability and in a timely manner are highly desired in our modern society. The RedBlade is a multi-functional autonomous vehicle which plows snow in the winter and mows grass in the summer. It is designed and implemented by a team of undergraduate students from multiple disciplines at Miami University. This vehicle competed and won second place in the Institute of Navigation´s 1st Annual Autonomous Snowplow Competition held during at the St. Paul Carnival in Minneapolis in January, 2011 and will enter the 7th Annual Autonomous Lawn Mower Competition in Dayton, OH in June, 2011. The RedBlade has a three-layer system architecture. The top layer is the navigation sensor suite. The current generation of the RedBlade navigation sensor suite includes a MEMS´s inertial sensor, a survey grade differential GPS receiver, a scanning laser, and optical wheel encoders. The middle layer is the collection of software that provides driver functions to the sensors and actuators, plow path plan, and vehicle motion control algorithm. The bottom layer is the mechanical platform, electronics hardware, including the motor controller, safety systems and power supplies, and processors that carrier out the software functions. RedBlade utilizes the navigation sensors to determine its position, heading, and velocity (PHV). The vehicle PHV information along with its predetermined destinations are fed to an onboard computer that implements a PID-based controlling algorithm to adjust the vehicle heading and velocity on the fly. We also included some additional heuristics within the control algorithm to help deal with slippage on icy road or wet grass and the weight of the collected snow on the plow. Both remote and on-board emergency kill switches allow an operator to bring the vehicle to a quick stop. This paper will present the detailed platform, navigation solutions, path planning, obstacle avoidance, and control algorithm design of the RedBlade and its performance evaluations as a snowplow and as a lawn mower.
[Return to Program]
Session A1, Paper #2
Lane Identification with Real Time Single Frequency Precise Point Positioning - A Kinematic Trial
R.J.P. van Bree, P.J. Buist, C.C.J.M Tiberius, Delft University of Technology, The Netherlands
Lane identification on a freeway may be required for next generation car navigation, and advanced driver assistance in general, as well as to support the observation and study of driver behaviour and traffic flow. These road vehicle applications call for sub-meter positioning accuracy, and some of them even for real-time operation, and all this preferably at low-cost. In this contribution we will investigate whether Real-Time Single Frequency Precise Point Positioning (RT SF-PPP) can meet the demands set by the above applications. Precise Point Positioning offers improved position accuracy, as compared to standalone GPS positioning, but without the need for local DGPS-like infrastructure.
The best position accuracy with SF-PPP is reached when precise GPS data products are used, i.e. final satellite clocks and orbits, final ionospheric maps and the latest Differential Code Biases. These products however are available to the user with a significant latency of a few days or even weeks after the measurement epoch, ruling out real-time operation.
At Delft University of Technology, a real-time version of SF-PPP has been developed. The SF-PPP algorithm uses un-differenced single frequency pseudorange code and carrier phase observations. The position solution is computed on an epoch by epoch basis (i.e. truly kinematic). For real-time operation, predicted satellite orbits, predicted Global Ionospheric Maps, and in particular real-time satellite clock estimates must be used. For the latter the RETICLE products from GSOC/DLR [3] are used, which have been shown to deliver comparable performance with IGS final products in [1]. The paper will also outline the requirements on data communication. In brief the required data rate to the vehicle on average is about 3 kbit/s, which is easily accommodated by today´s 2.5 and 3G telecommunication infrastructure.
Performance of Real Time Single Frequency Precise Point Positioning was extensively demonstrated in [1] and [2] for static applications. The position accuracy presented in [2] shows, a 95% error of about 0.30 meter in the horizontal directions, and 0.65 meter in the vertical. These results were obtained using high-end GPS receivers. For mid-end receivers these values were found to be still smaller than one meter in all directions. These findings have led to the question: can SF-PPP deliver the position performance required for lane identification of a road vehicle, in real-time, and preferably with low-cost equipment? Low-cost refers here to a simple single frequency GPS receiver, with a simple patch antenna, delivering pseudorange code and carrier phase observations, costing on the order of 100 US dollar or less.
To answer the research question a kinematic trial is performed with a vehicle on the road. The purpose of the trial is to assess the position accuracy and to test specifically the ability to identify the lane the car is driving in. Convergence of the PPP solution and data sampling rate are also subject of research.
For the test the antennas of low-cost and middle-cost receivers are all rigidly mounted on the test vehicle´s roof, next to several antennas of high-end equipment. The latter is used to reconstruct an accurate (cm-level) ground-truth for the low and mid-cost receivers, using differential carrier-phase GPS to a nearby reference station, namely a permanent GNSS station at TU Delft´s GNSS observatory, with accurately known position coordinates in the International Terrestrial Reference Frame (ITRF2005). In this way a highly accurate reference track is available for error analysis of the PPP-results. The multi-lane freeway (A13) used for the test is located between The Hague and Rotterdam, along Delft, and is one of the busiest freeways in the Netherlands. The test - carried out under ordinary driving conditions - is repeated several times during the day to evaluate possible effects of the ionosphere on the positioning results.
The position accuracy achieved with the low-cost receiver during the trial is about 0.30 m and 0.40 m, in terms of standard deviation for the horizontal and vertical coordinates respectively, with 95% error values of about 0.70 m and 1.0 m. These results show that RT SF-PPP can be used for lane identification. A lane on a freeway in the Netherlands is 3.50 m wide. Regarding the PPP convergence time after signal loss, the 68th percentile of the vertical position error reaches a level of 0.40 m in about 160 epochs. Correspondingly, the horizontal position solution reaches about 0.30 cm in 180 epochs. Depending on the data sampling rate of the receiver the convergence time after signal loss can take from 15-20 seconds to several minutes. Ionospheric conditions mostly affect the vertical component, and has only little effect on the lane identification.
[1] van Bree, R.J.P., C.C.J.M. Tiberius, and A. Hauschild (2009), Real Time Satellite Clocks in Single Frequency Precise Point Positioning, ION-GNSS-2009, Sept. 22-25, 2009, Savannah, USA, pp. 2400-2414. [2] van Bree, R.J.P., C.C.J.M. Tiberius (2011), Real Time Single Frequency Precise Point Positioning - accuracy assessment, submitted to GPS Solutions, 2011. [3] Hauschild, A., O. Montenbruck (2008), Real-time Clock Estimation for Precise Orbit Determination of LEO-Satellites, Proceedings of the ION GNSS Meeting 2008, Sept. 16-19, 2008, Savannah, Georgia, USA, pp. 581-589.
[Return to Program]
Session A1, Paper #3
A Comparative Study of Lidar and Camera-based Lane Departure Warning Systems
Jordan Britt, C. Rose, D. Bevly, Auburn University
Recent work in vehicle safety systems has been focused on augmenting GPS-based systems with other sensors to provide solutions even when GPS is unavailable such as in tunnels, dense foliage, or urban canyons. However, these additional sensors have their own rates of failure. Knowledge of the conditions in which each sensor can fail is important for robust vehicle safety applications. One such obvious application of this is in vehicle safety systems, specifically, lane departure warning (LDW) systems.
Lane Departure Warning systems provide an effective means for preventing accidents on highways due to driver inattention, sleepiness, and other factors. Historically, there have been two main types of sensors used to detect the lane markings, first Light Detection and Ranging (LiDAR) sensors use reflectivity for detection and Camera-based lane departure warning systems which extract visual information from the scene. Both sensors performance, however, can be hindered due to various environmental conditions that can lead to false alarms or misses. This paper will discusses the various strengths and weaknesses of two prototype camera-based and LiDAR-based lane departure warning systems in an effort to determine in what scenarios does one sensor out perform another.
LDW systems based on camera vision have been extensively researched and are present in vehicle safety systems. The various methods for detection and tracking of lane markings on the road vary widely. For this paper, a prototype camera-based LDW system was designed which uses a dynamic threshold to build a binary image of the road. Edge detection extracts the edges of the image, and the Hough transform is used to find lines. Each of these lines are then processed through two selection criteria, a slope criteria and a location criteria, to eliminate lines from the image which do not correspond to the lane markings. The valid lines are grouped into left and right lane markings, and the endpoints and midpoint of each line are collected into left and right lane marking pools. Each pool undergoes a least squares polynomial interpolation to build a 2nd order polynomial model of the lane marking in image space. The coefficients of the 2nd order polynomial model are used as measurements in a Kalman filter to further reduce errors in the system. Lateral distance from the lane markings is calculated using the distance in pixels from the previously-defined center of the vehicle in the image and the lane marking model. This distance in pixels can then be converted to real world distance using a known conversion factor. Lidar based LDW systems have been an area of significant research. While a host of different techniques exist, this paper will concern itself with the use of a multi-layer lidar to detect lane markings by fitting the scan data to an ideal model of the lane´s reflectivity. This ideal lane model is essentially an idealized representation of the lane´s reflectivity signature, where the road surface is modeled as a const area of reflectivity bordered by spikes in reflectivity representing the lane markings. This lane model is then fit to the actual lane reflectivity data by minimizing a mean square error in an exhaustive search over a minimum and maximum lane width. Once the lane markings are detected, a narrowed search area is created to ease future detection. Lateral distance from the center of the lane is calculated and filtered to minimize erroneous jumps in the data.
Each LDW system will be tested in various environmental conditions to ascertain the strengths and weaknesses of each system. The conditions include differences in light such as dawn, day, and night. Road lane marking types such as dashed, double, yellow, or white will also be examined. Finally, testing will determine the performance of each system in weather conditions such as rain. Emphasis of the analysis will be placed on the comparison of each sensor. The true position of the vehicle in the lane during these tests will be determined from RTK GPS and a precision survey of the lane markings at the NCAT (National Center of Asphalt Technology) test track. It is from this lateral position in the lane that the algorithms will be compared. This comparative analysis will specifically determine the situations in which each sensor fails and quantify the performance of each sensor.
[Return to Program]
Session A1, Paper #4
Air Target Detection Using Navigation Receivers Based on GPS L5 Signals
V. Behar, Institute of Information and Communication Technologies, Bulgaria; C. Kabakchiev, Sofia University, Bulgaria; H. Rohling, Technical University Hamburg-Harburg, Germany
The secondary application of navigation receivers can be in Bistatic Radar (BS) with a non-cooperative transmitter. Such BS is a specific case of passive radars that exploits broadcast and communications signals as ´illuminators of opportunity´. Such systems have very useful features. In difference with ground-based BS systems, where the transmitter and receiver can be designed to operate together, in this paper we consider a BS radar configuration, in which GPS is used as a space-based transmitter. This means that the receiver has no knowledge of the transmitted signal and the direct signal from GPS is used for synchronization between transmitter and receiver. The feasibility of using GPS as a transmitter of opportunity is firstly examined for the case of BS radar, whose power budget is approximately evaluated for the only civil GPS L1 signal (GPS C/A). The civil L1 signal is transmitted by satellites at 1572.42 MHz and contains the coarse acquisition (C/A) code, which is unique for each satellite. The C/A code modulated signal is a BPSK signal with a chip rate of 1.023 MHz and the repetition interval of 1ms. The L1 signal frequency bandwidth is 2.046 MHz. The power budget analysis of such a BS system shows that due to the extreme weakness of the transmitted GPS signal L1, detection of air targets with small RCS is still impossible. However, modernization of GPS, which is now underway, provides an opportunity to use the improved properties of some new designed civil GPS signals (for example, L5) in BS radar, which exploits GPS as a transmitter of opportunity. The GPS L5 signal is designed to support safety-of-line applications such as aviation navigation, and the major innovations brought by GPS L5 signal, with respect to GPS L1 signal, are the additional NH code modulation and the use of a pilot channel free of data. The L5 signal is transmitted at 1176.45 MHz with a received power of -154 dBw, which makes the L5 signal four times stronger than the L1 signal. Therefore, the bandwidth of the L5 signal is increased to 20.46 MHz, which is ten times wider than the bandwidth of the L1 signal. The I5 and Q5 components are then modulated by a 10-bit NH-sequence and a 20-bit NH-sequence, respectively. Each bit of the NH-sequences is 1ms, resulting in 10ms period of the I5 component and 20ms period of the Q5 component, respectively. In this paper we calculate and analyze the probability characteristics (detection and false alarm) of a BS radar system for air target detection, which consists of a GPS satellite used as a transmitter and a receiver located on the earth´s surface . The operation of such a BS radar system is based on the following principles: the position of a transmitter varies over time but is predictable with some accuracy; the transmitter antenna illuminates all the sector of target detection with the GPS L5 signal; the receiver performs fine separation of the direct signal (DS) from a satellite and the potential target signal by using optimum beamforming; the reference signal needed for the correlation performance is obtained by synchronization of NH20 code in the incoming L5 signal from GPS ; In several our paprers we use the same investigation approuch for calculate and analyze the probability characteristics (detection and false alarm) of the coarse acquisition (C/A) code. In the input part of a receiver the received signal in each antenna element is down converted and converted to digital. Then the adaptive beamforming system, which uses, for example, the Minimum Variance Distortionless Response (MVDR) algorithm, creates minimally two beams. The fist of the created beams has a deep null in the direction of a satellite in order to suppress the direct path signal (the target channel), and the second one has the maximum gain in the satellite direction (the reference channel). The reference signal needed for cross-correlation is obtained by synchronization of the NH20 code in the reference channel. The synchronization process consists of two parts - acquisition and tracking. A 20ms NH code can be used for acquisition. The coarse frequency estimate obtained using the acquisition algorithm is then passed to the tracking algorithm for fine Doppler estimation. The coherent processing is performed by correlating the reference signal with the frequency shifted target signal. After the additional non-coherent integration the signal detector using CFAR (Constant False Alarm Rate) algorithms can be applied in order to indicate detection of a target.
This paper will present theoretical and numerical results for air target detection using the secondary application of the GPS L5 signal. For comparison two variants of a BS radar system are considered. The first of them is concerned with the case when the bistatic angle ? is smaller than 140ยง. This is the classical variant of BS radar. The second variant of a BS system is forward scattering radar (FSR). It is the case when the target is located close to the baseline and the target RCS is significantly enhanced compared to the typical values. The detection characteristics are calculated under the following assumption: target detection is carried out on the background of a white Gaussian noise; target - low-flying and poorly maneuverable (for example, helicopters); signal - Q5 component of the GPS L5 signal; in the forward scattering region - the antenna array adaptively creates very narrowband beams in the direction of both the satellite and the target. This means that the signals received from a satellite and a target are still separable. The numerical and graphical results that prove all the theoretical calculation will be presented. The comparison analysis of the results obtained will be made.
[Return to Program]
Session A1, Paper #5
GNSS Interference Monitoring Network Based on Detection in Automotive ITS Station Receivers
R. Bauernfeind, I. Kramer, B. Eissfeller, University FAF Munich, Germany
GNSS is an essential sensor for location based services, intelligent vehicle technologies and intelligent transport infrastructure. GNSS signals are highly vulnerable to interference no matter if intentional or unintentional. There are several sources causing unintentional interference in GNSS frequency bands but the most serious threat is intentional interference coming from so called InCar-Jammers. InCar-Jammers block GNSS signal reception in their vicinity and degrade GNSS performance, proportional to the distance, over a wide area. They are relatively easy to purchase from abroad over the internet and operated by plugging them into the cigarette lighter of a vehicle. The motivation of using an InCar-Jammer can be prevention of being tracked by a fleet management or theft protection system or a fraud attempt against a distance based charging system, etc. A national infrastructure for detecting and locating GNSS interference, to protect Intelligent Transport Systems (ITSs) and their revenue streams, is needed. Emerging vehicular communication is an enabling technology for future ITS applications. It requires knowledge about the current position for geographical information dissemination, i.e. GeoNetworking. Clearly, degrading the integrity of GNSS positioning is a threat for all safety relevant ITS applications. Therefore, avoidance and mitigation of interference signals is subject of safety related vehicular communication and the ITS standards should be able to handle this in the same way as other safety related issues. The introduction of a GNSS interference monitoring capability in ITS station receivers would establish a comprehensive national interference monitoring network and should therefore be considered within the standardization process of vehicular communication.
First the paper analyses intentional interference transmitted by InCar-Jammers, but also unintentional interference sources will be considered. To analyze intentional interference sources, various InCar-Jammers have been purchased. The signal of the InCar-Jammers and their effect on the GNSS receiver has been analyzed by open field interference tests in the Galileo Testbed (GATE) in Berchtesgaden. At the Galileo Testbed different scenarios have been simulated e.g. with a stationary jammer at the roadside and a jammer moving with a vehicle. The area around the InCar-Jammer where the receiver front-ends are saturated, the receivers loses lock on the tracking loops and where the positioning accuracy is degraded are analyzed. InCar-Jammer Signals have been recorded with the ipexSR Software Receiver for further analysis and evaluation of detection algorithms. The ipexSR Software Receiver is a real time capable multi frequency receiver, developed at our institute. An overview on the available InCar-Jammers with an analysis of their signal characteristics will be presented in the paper. For unintentional interference, sources like malfunctions in transmitters generating harmonics in the band of GNSS or interference in the L5/E5 band through aeronautical radio navigation aids as DME will be considered.
The second part of the paper covers detection algorithms required to establish a GNSS interference monitoring network with the ability to identify and characterize the interference source within the receiver. Different techniques of analyzing the signal in the time-frequency domain will be presented. Algorithms to detect and characterize the InCar-Jammer interference will be implemented and tested with the Institute´s ipexSR Software Receiver. The performance of time-frequency domain transformation (e.g. Short-Time-Fourier-Transformation, Wavelet-Transformation) as well as the detection algorithm and algorithm to characterize the interference source will be given.
The third part covers the interference monitoring network architecture based on the vehicular communication architecture. Detected interference shall be communicated to warn other vehicles about interference in their vicinity and to a back-end where the interference shall be localized. Advantages of additional sensors and map mapping available in vehicles to assist the localization are discussed. The detection of an interference event by ITS station receiver triggers the emission of a decentralized environmental notification message (DENM) containing characteristics about the interference source. This message is sent via vehicular communication to the backend, i.e. local authorities like for example the responsible traffic management center (TMC). TMCs have a close link to police forces which can get active then. Messages related to the same interference source are evaluated at the TMC and an enriched DENM is assembled containing all available information about the interference source. The assembled message is then broadcast within the relevant area via GeoBroadcast. Localization of the interference can be done at a back-end office where the measurements from various OBUs are collected through algorithms based on DRSS (Differences of Received Signal Strength) of the measured interference power.
With the introduction of an open service signal on L5/E5, automotive receivers will take advantage of this signal which then will also cause future InCar-Jammer to operate in the L5/E5 band. InCar-Jammers operating in the E5 band will not only interfere with the GNSS service but also with other aeronautical radio navigation services. In such an environment it would be even more necessary to operate a comprehensive GNSS interference monitoring network. The implementation of such a system directly at the user´s automotive receivers has the advantage that the detection is done directly at the threatened applications which also can take immediate counter measurements. A common introduction with vehicular communication has also advantage because both systems have requirements on the coverage to enable specific applications which will enforce a fast introduction by the automotive industry.
[Return to Program]
Session A1, Paper #7
GNSS Shareware Control Software for Droid and Apple Apps
S.D. Lyle, R.A. Smith, Texas A&M University Corpus Christi
New compact multiple channel/frequency OEM board GNSS receivers are being developed with the capability to meet some key critical hurdles for mass market utilization which include cost, size, and power. Control software to interface these systems are built for PC or proprietary hardware systems. The control software interfaces with the GNSS receivers execute simple and advanced functionality. Mobile cellular phones have the ability to support control applications of GNSS receiver while utilizing the internal cellular existing GPS unit for augmentation for the external GNSS receiver. These GNSS receivers often come with an API that can be used to quickly build a mobile graphical field solution. The GNSS receiver can be controlled via Bluetooth by mobile devices for raw data collection and real-time solutions by passing DGPS or RTK corrections. GNSS controller software applications can be built on mobile cellular devices that can be shared on Windows Mobile, Droid, or Apple platforms. This research will discuss a software shareware ´App´ product on a mobile cellular phone that can interface with a compact GNSS receiver. The results will discuss of the methods used develop software for use in mapping and surveying applications. The process of augmenting the GNSS OEM receiver and the mobile cellular GPS receiver to quickly obtain RTK solutions will be presented. A demonstration of how the combination of the mobile cellular GPS, GNSS OEM receiver and the shareware software ´App´ can be used for hydrographic, aerial platform, GIS, emergency response, and surveying/engineering applications is presented.
[Return to Program]
Session A1, Paper #8
Automotive Urban Assist Enabled by Satellite Navigation Technology
E. Wasle, P. Berglez, TeleConsult Austria GmbH, Austria; C. Mongredien,
Fraunhofer Institute for Integrated Circuits IIS, Germany; A. Kahmann, OECON Products & Services GmbH, Germany
The primary objective of advanced driver assistant systems (ADAS) is to disburden the driver in complex traffic situations, thereby mainly increasing the safety of the driver and other road users and reducing the number of accidents. Automotive Urban Assist in particular aims to significantly reduce the number of urban road accidents and consequently the number of injured and killed people in road traffic. As a secondary effect, ADAS will help to reduce traffic jams, resource consumption, air pollution and CO2-emissions.
Satellite navigation receivers are increasingly used as an additional sensor within different kinds of ADAS. At first GPS was used for in-vehicle navigation systems only, but the introduction of satellite navigation in safety relevant systems is becoming suitable with the ongoing modernization in satellite navigation. The increasing number of civil signals, their signal design, and their separation in the frequency spectrum increase the interference resistance, the availability, and the redundancy. Furthermore multi-frequency measurements are used to correct ionospheric effects, and mitigate multipath. Another driving factor for using satellite navigation in new application fields is the increasing number of satellites when jointly using compatible GNSS. Accuracy, availability, but above all reliability and integrity are important performance measures which have to be met by GNSS receivers in order to find their way into ADAS.
Consider for the moment two particular road traffic scenarios where ADAS can support the driver: firstly collision avoidance during a "left-turn" (right-hand traffic considered); and secondly a "stop-line" recognition, to avoid accidents e.g. through a red light violation warning. Various situations have to be differentiated within these two scenarios, depending whether the own car is moving and with what speed; whether the oncoming traffic is moving; whether there are traffic lights; how many lanes there are; what other road users are in the intersection; and so on and so forth.
Any ADAS has to perceive the environment and the traffic situation as it would be perceived by a driver. The static situation consists of the number of lanes, the driving direction, the existence of traffic lights, etc. This can be provided in form of attributed maps, which have to have sufficient accuracy in order to clearly identify lanes, or the position of stop-lines. Stopping one meter behind the real stop-line is not an option for a safety relevant ADAS. The dynamic situation, thus the oncoming traffic, the status of the traffic light, or the own position, has to be determined in real-time. The interaction of various sensors is needed in order to provide all necessary information. Thereby, a certain level of redundancy increases the level of integrity.
The GNSS sensor provides the absolute position, velocity, and time reference for the integrated system, which is additionally connected to the existing on-board electronics and standard diagnostic. The GNSS sensor, moreover, is the only mean to relate the static situation stored in maps with the dynamic situation perceived. This makes implementing GNSS in ADAS systems interesting and new. The integrity for the satellite signals will be provided by the wide area augmentation systems or the satellite navigation inherent integrity facility. For Europe in particular of interest is the combined use of GPS and EGNOS on the one hand and Galileo on the other. This will provide a high level of integrity and availability even in urban canyons. In addition, car sensors will be integrated and autonomous integrity monitoring algorithms applied to provide timely alerts about any deterioration of the navigation solution. In this way the position, velocity and time of the own car is determined. The oncoming traffic and stop-lines are perceived using radar, lidar or visual stereoscopic sensors. The exchange of position information between cars is beyond the scope of this paper.
In the first scenario of "left-turn" the system has to provide a warning if oncoming traffic may collide with the own vehicle. It is important to allocate all perceived objects and the own vehicle to a specific lane. This is only possible with a reliable localization of the car and a precise map of the intersection. In the "stop-line" scenario, the driver will get a warning, if the driver is going to interfere with a stop-line. Therefore the position and velocity of the own car has to be known, as well as the location of the stop-line. The stereoscopic sensor can support the system as long as the stop-line is clearly visible and not hidden by slush. A stop-line in combination with a traffic light additionally requires status information from the traffic light.
A number of different sensors have to be developed and enhanced in order to provide the necessary functionality for the envisioned Automotive Urban Assist scenarios. Thereby central element is the development of a scalable, inexpensive high accuracy positioning system for urban environments with built-in integrity monitoring. The envisioned GNSS receiver shall receive GPS, Galileo and EGNOS signals on L1/E1 and L5/E5 frequencies and incorporate assistance and differential information from a remote server. EGNOS messages, ephemeris data, approximate user position, differential data of a differential network or of a local reference station will be transmitted through wireless communication links to the receiver. The server thereby applies intelligent selection strategies to provide a minimal data set with maximal performance to use only small communication bandwidth and costs. The positioning unit fuses the vehicle sensors with the satellite navigation message to a position, velocity, and time information whereby any faults and failures in the measurements have to be detected and excluded.
The research work described is conducted within the European GNSS Agency funded project "Galileo / EGNOS Enhanced Driver Assistance" (GENEVA) with the final objective to support the driver in critical situations in particular in urban intersections.
This paper introduces the GENEVA project and specifies the Automotive Urban Assist application and highlights the selected scenarios under test. Further it describes the automotive, multi-frequency GNSS receiver and its positioning software which both are under development, and the assistance and differential data server which provides all required data via a safe and reliable wireless communication link.
[Return to Program]
Session A1, Alternate #1
A GPS/Galileo Tightly-coupled Localization System for Safety-relevant Automotive Assistance Systems
H-G. Busing, M. Escher, T. Scheide, P. Hecker, Technische Universitat Braunschweig, Germany
The German Federal Ministry of Economics and Technology launched a project to evaluate the use of GNSS based localization systems for safety relevant driver assistance systems. The project "FAMOS - Galileo for Future Automotive Systems" is led by the Volkswagen corporation, the Institute of Flight Guidance is developing a localization module for three safety relevant driver assistance systems. The applications that have been defined in the project have their main focus on green driving and safety improvement.
The Institute of Flight Guidance has presented a system architecture for an automotive localization system at the ION GNSS 2010. Based on a series of test drives in different environments the system has been defined. The localization unit makes use of three sources of input: GNSS raw data measurements, data from the CAN bus for dead reckoning as well as status information from the vehicle controllers. In the meantime the system has been implemented and the performance will be presented in the proposed paper.
The localization unit has been implemented in the "Automotive Data and Time-Triggered Framework (ADTF)". It is a software framework that allows processing data from various input sources. The framework cares for synchronous data handling, allows real-time data processing and provides a data recording and playback utility.
As a development platform, two test vehicles are operated in the FAMOS project: The Institute of Flight Guidance operates a Volkswagen Passat station wagon. This vehicle is primarily equipped for research in the positioning domain. Non-standard hardware such as a high-precision inertial measurement unit, embedded PCs and additional antennas for GNSS measurements and data transmission are installed. This vehicle is primarily used for the development of the localization system. Additionally, a Volkswagen Golf VI is concurrently being set up with additional sensor devices to provide a basis for perception of the surrounding traffic.
For the development of the localization data fusion, commercial GNSS receivers from Novatel (OEMV and OEM6) are being used for GNSS raw data measurements. CAN data from the vehicle sensor network is streamed to the processing computers via a Vector CANboardXL device. As the measurements from the vehicle sensors do not have any timing information, they are time stamped with the GPS time from the Novatel GNSS receivers. The time stamping device has been developed by research engineers of the Institute of Flight Guidance.
The localization system is realized as a tightly-coupled data fusion. In each update step, the vehicle state is predicted based on the kinematic model based on the Ackermann geometry. The measurement update is then performed based on the available data type. If the data fusion receives data from the vehicle network, a dead reckoning update will be performed with measurements from the rear wheel odometers, a longitudinal accelerometer and a yaw gyroscope. These sensors are standard series equipment of the used vehicle. If the measurement data is received from the GNSS receiver the vehicle state will be updated using GNSS range and Doppler measurements.
As the Galileo signal in space will not be available during the FAMOS project, development and testing of the system is carried out in the Galileo testbed "aviationGATE" that the Institute of Flight Guidance operates at the research airport of Braunschweig. The "aviationGATE" comprises nine Galileo E1/E5 pseudolites that are positioned directly at the airport or on surrounding hills. The localization unit is able to handle both GPS only measurement updates as well as hybrid GPS/aviationGate measurements. Although the system time of "aviationGATE" is adjusted to the GPS time, the data fusion does consider potential timing errors.
Due to the limited availability of additional open civil GNSS frequencies, mainly the L1/E1 signal will be used for positioning in the FAMOS project. Where available, a second frequency is used to monitor ionospheric delay.
As the positioning module is used as a base system for safety-relevant driver assistance systems, the integrity of the positioning solution is a major requirement towards the data fusion. SBAS is used to obtain integrity information for the signal in space. Due to the low elevation angles of the geostationary satellites and limited availability of the signal in rural and urban environments, the SBAS messages will be streamed from a ground-based EGNOS data access point. Pseudorange corrections for differential positioning are received from a nationwide reference station network called SAPOS. It has been indicated in previous publications from the authors that the precision of positioning can be improved compared to SBAS corrections due to a denser network of reference stations. While SBAS is considered a regional augmentation system, the SAPOS service provides local range corrections.
Due to the fact that measurements both from the GNSS receiver as well as from the vehicle sensors may suffer from disturbances, the integrity of the input sources is monitored. One of the most likely sources of errors in the GNSS signals are multipath effects. Especially in urban environments, signal reflections from surrounding buildings are likely to reach the antenna. The localization module checks the data sources´ integrity in a two step approach: Pre-fusion integrity checks and in-fusion integrity checks are implemented. Pre-fusion integrity checks make use of redundancies by excluding faulty measurements based on plausibility checks with the kinematic model. In-fusion integrity checks include measurements that are unlikely to be reasonable based on residual checks.
[Return to Program]
Session A1, Alternate #2
Accurate Relative Localization for Land Vehicles with SBAS Corrected GPS/INS Integration and V2V Communication
M. Obst, E. Richter, G. Wanielik, Chemnitz University of Technology, Germany
Accurate and reliable positioning for vehicles in automotive applications is still an unsolved problem. The requirements in terms of performance and efficiency are steadily increasing. While modern safety and traffic management applications depend on lane level accuracy, standalone GNSS receivers are not able to continuously provide them. In addition to this, absolute knowledge of the position of a vehicle is not always needed: e.g. for ADAS applications like blind spot detection only a relative distance between vehicles is necessary. In contrast to introducing dedicated ranging sensors like Lidar or vision based systems, the usage of GNSS observations and vehicle-to-vehicle (V2V) communication could be considered. The main motivation for applying relative localization is, that nearby GNSS receivers are normally subject to identical kind and magnitude of errors (e.g. same tropospheric and ionospheric delay). Throughout the process of relative localization these errors cancel out and the resulting relative vector should be more accurate than the corresponding absolution positioning solution. In this paper a method for accurately estimating the relative position between an arbitrary number of vehicles by using low-cost standard GPS receivers is presented. Since vehicles are normally not directly connected, the GNSS observation data is exchanged via ad-hoc 802.11p V2V communication channels. For the proposed system, each vehicle is equipped with a standard single-frequency GPS receiver which is capable of delivering GNSS raw data (pseudoranges) at a reasonable updated rate like 5 Hz. These GNSS observations are continuously broadcasted and thus shared with nearby vehicles. Finally, the relative vectors between the ego vehicle and the surrounding vehicles are estimated through an implementation of the Bayesian Filtering Framework. Furthermore, for vehicular applications, a physical model which accurately describes the constrained vehicle´s motion can be applied. Since modern cars are equipped with in-vehicle odometer sensors which provide wheel speed and yaw rate measurements, this information can be directly used by an appropriate vehicular motion model. By that approach a tightly coupled GPS/INS system which additionally considers GNSS raw measurements from remote vehicles is described. In order to further improve the estimation of the pseudorange error components Europe´s Satellite Based Augmentation System (SBAS) implementation EGNOS is utilized. In literature there exists a variety of different algorithms for performing relative localization with the aid of GNSS localization. One straightforward approach can be described by simply deriving the relative vector from differencing two absolute GNSS fixes. While this method works well under ideal open-sky conditions, it immediately breaks if the satellites in view of both vehicles are different. In this case, the initial assumption that both receiver solutions suffer from the same local errors is violated and the resulting relative vector is inaccurate. An improved version of the prior described algorithm would not directly use the positioning solution from the receivers. Instead the position fixes would be calculated from the raw pseudoranges while ensuring that the positioning solutions are generated from a common subset of visible satellites for each vehicle. Even though this method guarantees a consistent relative vector, the main drawback of not always using all available GNSS measurements should be considered. Another well-known technique for applying relative localization is through the use of double differencing. Here the raw pseudoranges from one epoch of two independent GNSS receivers and satellites are subtracted. As a result the satellite and receiver clock errors are cancelled out. Additionally, for a short base line distances between two receives-which is true for the proposed system-the tropospheric and ionospheric delays are eliminated as well. As indicated above, the pseudorange measurements have to originate from the same epoch which could be challenging in a decentralized loosely organized environment. Furthermore, an incorporation of an advanced pseudorange error model is not possible with the outlined approach. The relative localization algorithm presented in this work is based on a probabilistic estimation process rigorously utilizing Bayesian Filtering. The aim of the Bayesian algorithm is to recursively estimate the probability density function (PDF) of a system´s n-dimensional state space at each time step k. The vehicle´s dynamic behaviour is described by a discrete-time stochastic model (vehicular motion model). The state vector of the outlined system consists of vehicle´s pose and the typical dynamic parameters like velocity and turn rate. It further includes the clock bias and clock drift of the in-vehicle GPS receiver. In order to derive an accurate relative vector the pseudorange errors for each satellite in view are also included in the estimation process. If an unknown remote vehicle enters the V2V communication range the state space is dynamically augmented with a new pair of position states. Raw GPS observations from the local receiver as well as the odometer measurements from the vehicle´s CAN bus are directly applied through two individual measurement models to the state variables of the ego vehicle. The remotely received pseudoranges are only updating the prior augmented position states for the corresponding vehicle. Both, the local generated and the remotely received pseudoranges are further rectified by the EGNOS corrections (User Differential Range Error and Grid Ionospheric Vertical Error are handled in the likelihood function) received via signal in space or the EGNOS Data Access Service. Finally, the generation of the relative vector can by simply done by differencing the x,y state variables of the ego vehicle with the x,y states of the remote vehicle at any time. The relative localization algorithm described, will be tested and validated through real data recorded with two research vehicles available at the authors´ university. Both vehicles are equipped with high reliable reference sensor systems (Novatel SPAN System with RTK and IMU). Furthermore the results will be evaluated against already existing (see above) relative localization algorithms. The authors expect an improved relative localization performance both, in terms of accuracy and consistency. Especially for challenging environments where for example not all satellites are visible by each vehicle, a robust solution should be provided. Furthermore, through the stabilization of the estimation process with a vehicular motion model and odometer sensors, possible GNSS observation outliners should be mitigated.
[Return to Program]