In addition to communication and positioning capability, The Lane Change Advisor feature monitors traffic in
GM V2V demo vehicles are equipped with various lanes adjacent to the host vehicle, and displays
displays and actuators. These include High Head-Down information to the driver when a communicating vehicle
Displays (HHDDs), mirror icons, haptic seats, and certain is detected in the host vehicle blind spot. This feature also
brake controls. The indicators are mostly application alerts the driver when a lane change maneuver is unsafe
specific and some, such as the HHDD, are used in due to a passing vehicle equipped with the technology.
multiple applications. Figure 2 shows the mirror icon The intention of this feature is to notify the driver when
indicating the presence of a vehicle in the blind spot of there are vehicles in adjacent lanes that the driver may
the host vehicle (left), and the HHDD icon indicating the have difficulty seeing. A display embedded in the side
presence, caution, and warning (green, amber, and red, mirror illuminates an icon to warn the driver of the
respectively) generated by a vehicle in front of the host presence of a vehicle in the host vehicle blind spot.
(right).
• Configuration information for reference and low- The error checking method adopted for the L1/L2 truth
cost receivers solution in this test platform used two separate antennas
• Data logging requests and format information and high-end receivers, and checked to see if the distance
• Port information of requests to GPS data between them as estimated by RTK GPS was consistent
connections with the physical distance between them. The RTK
• Serial connection information for receivers solution for each receiver was estimated, and based on
these solutions, the distance between the two antennas
was calculated. If the calculated distance-between-
GSG Overview antennas significantly differed from the known distance
compared to a predefined threshold, both solutions were
The GSG received data from two RMs by connecting to rejected. Additionally, the elevation angle was checked to
an RM as a client. GSG is designed such that, in full-scale ensure that it was a reasonable value.
implementation, it will be able to connect to RMs in all
equipped neighboring vehicles. GSG generates near-real- Figure 4 illustrates the reasoning for using a two-
time relative position and velocity estimates from the GPS antenna/two-receiver setup rather than using the ratio test
data. GSG algorithms are capable of operating in single- to determine the correct estimates. The driving route for
frequency or dual-frequency mode and are user this data was through suburban and commercial areas
configurable. GSG estimates are transmitted via a socket ranging from fairly clear to heavily canopied with trees,
interface to user computers. If one of the RMs supplies with a short segment of freeway driving. Overall, 80% of
additional high-accuracy truth data, the GSG is able to epochs had four or more satellites in lock. Figure 4
provide metrics for accuracy. Configuration of the GSG depicts the cumulative probability for the ratio for the
includes the following: epochs determined to be correct (accepted) by the two-
antenna system, as well as the cumulative probability
• IP numbers and port numbers for connecting distribution for those determined to be incorrect
RMs (rejected). If a ratio test was used at a typical 0.3 value,
• Local RM (only the local can process truth data) 2.1% of epochs would be in error. More importantly, only
29% of the correct solutions would have been accepted.
• Local RM reference data availability
Using the Ambiguity Dilution of Precision (ADOP) at a
• Single- or dual-frequency data availability
level of 11, 2% of epochs would be in error, and a slightly
• Logging information for GSG estimates
better 36% of the correct epochs would be accepted. With
the two-antenna test, 81% of solutions were accepted as
Recorded data from all connected RMs and GSGs can be
correct.
processed to obtain the same type of output as the near-
real-time system. The post-processed low-cost data can be
It must be noted that, for these tests, obscuration was a
used to evaluate the performance of the low-cost GPS
much more serious problem than it would be for a single
hardware.
vehicle: both vehicles must simultaneously have an
adequate number of GPS satellites visible. We have not
yet analyzed the gap bridging time requirements for the
Reference Solution Generation case in which both vehicles determine their position using
an external reference station.
One of the main problems with RTK GPS is the
validation of RTK estimates with a high degree of
reliability. Low residuals and a low value of the ratio of
the residuals of the best solution to the second best
solution are widely used for such validations [Misra and
Enge, 2001]. However, these tests discard many good
solutions.
Several solution types were considered for the GSG. The GSG PERFORMANCE EVALUATION
following review outlines the considerations involved in
designing the final GSG solution procedure. Performance evaluation involved initial bench testing and
validation of the system to low-cost GPS evaluation tests
A single-frequency, single-epoch least-squares relative in different test environments.
solution was considered. In principle, with sufficient time
in lock, such a solution would provide sub-meter
horizontal accuracy. However, part of the design goal was Initial Testing / Filter Tuning
to provide a solution capable of decimeter-level relative
positioning accuracy, and it seemed that least-squares Bench testing consisted of stationary testing to debug
relative solutions would not be capable of fulfilling that communication issues. Timing routines were
requirement. Nonetheless, straight-forward Hatch filtering implemented to estimate latency in the system. It was
[Misra and Enge 2001] with carefully chosen dynamic found that a fully loaded GSG (one that computes one L1
weighting and a single-epoch least squares solution filtered solution and two L1/L2 fixed solutions) runs in
should not be discounted for such applications. about 100 milliseconds, approximately half the time used
by Lambda algorithms.
SRI has considerable experience in dual-frequency single-
epoch fixed integer solutions, and an attempt was made to Initial dynamic tests began with the rover and reference
use similar methodology in the single-frequency case configured on the same vehicle, with a known antenna
[Sinko and Strus 2002 and Sinko 2003]. However, single- separation. With a fixed, known relative distance,
frequency ambiguity resolution is much less reliable than preliminary filter tuning was simplified. This filter tuning
its dual-frequency counterpart. As such, single-frequency phase was conducted only with the high-cost receiver. As
applications are typically used only over multiple epochs an example scenario, the test vehicle was driven into an
and on short and known baselines, such as in attitude open area near the San Francisco Bay. The test area
determination systems. However, an attempt was made to provided open sky access down to an elevation angle of 5
provide unconstrained, single-frequency, single-epoch degrees. The system was powered up as the vehicle was
solutions for this project. This technique produced usable driven at coasting speeds. In post-processing, the true
solutions for less than half the total number of epochs, inter-antenna distance was subtracted from the estimated
even under ideal conditions. As such, this type of solution distance calculated by the GSG. The results were used to
proved too risky for automotive safety applications. estimate filter uncertainty parameters.
Another solution was considered with the inclusion of Figure 5 indicates the true error and the filter standard
vehicle dynamics. Since the vehicle type is known, it was deviation. In general, the filter is tuned pessimistically by
assumed that the relative dynamics could be modeled. design. The initial standard deviation depends upon the
Therefore, it made sense to incorporate vehicle states into exact constellation at the time. However, the figure results
a filter and recursively estimate float ambiguities. Similar are fairly typical. In this case, the filter range standard
attempts were found in research literature, and deviation starts at about 2.3 meters and converges to 0.5
considerable room for improvement was noticed in
Integer Fixing
Figure 5. True range error (blue) and filter uncertainty of
range error (red). In addition to the single-frequency filter solution, the
fixed integer solution for the high-cost receiver was
computed, as illustrated in Figure 7.
Menlo Park Tests
Fixed
High 0.00 0.054 98.68 0.356
Integer
The most significant ongoing work is focused on De Jonge, P., and Tiberius, C. [1996] The LAMBDA
improving reliability and availability of positioning Method for Integer Ambiguity Estimation:
systems in all environments, especially urban canyons. Implementation Aspects, LGR-Series No. 12,
While significant improvements have been implemented Publications of the Delft Computing Centre.
in vehicle platforms over the years, there is ongoing work
to make these systems even more reliable and available, Duffy, J. [2005] GM to Rollout Intelligent Car
such that these functionalities can be deployed in mass Alternative, Network World, November 21.
scale. General Motors is currently working on integrating
various in-vehicle sensors with GPS to further improve Ford, T., Neumann, J. [1994] Novatel’s RT20-A Real
system reliability and availability, particularly in urban Time Floating Ambiguity Positioning System,
environments. This effort includes investigating new Institute of Navigation’s ION GPS 1994.
sensor technologies, as well as integrating latest
technology enhancements to existing systems. General Ford, T., Neumann, J., Fenton, P., Bobye, M., and
Motors is also investigating the use of modernized GPS Hamilton, J. [2001] OEM4 Inertial: A Tightly
and Galileo GNSS in future vehicle platforms. Future Integrated Decentralized Inertial/GPS Navigation
GNSS developments are expected to further improve the System, Institute of Navigation GPS Meeting 2001,
reliability and availability of GNSS-based positioning. Salt Lake City, UT, Sept. 11–14.
While this work is focused on road-level and lane-level General Motors and Delphi-Delco Electronic Systems
positioning capability, General Motors is also [1998] Final Report: Automotive Collision
investigating in-lane-level positioning capability for long- Avoidance System (ACAS) Program, National
term applications. New algorithms and processes are Highway Traffic Safety Administration, U.S.
being investigated using existing and future GNSS Department of Transportation.
configurations.
Ghoneim, Y., Lin W., Chin, Y. K., Sidlosky, D. [2005]
Enhanced Traction Stability Control System, SAE-
ACKNOWLEDGMENTS 2005-01-1591, SAE World Congress, Detroit, MI.
The authors greatly appreciate the contribution of Mark Kellum, C. C. [2005] Basic Feasibility of GPS
Dolan at SRI International who developed the RM Positioning without Carrier-Phase Measurements as a
component of the relative positioning test platform. Relative Position Sensor between Two Vehicles,
Institute of Navigation’s NTM-2005.