GM, NHTSA Move into Phase II of ACAS Forward Collision Warning Project
IVsource.net
27 February 2001

After extensive development and testing, GM and government engineers have certified a functional prototype for the Advanced Collision Avoidance System (ACAS) -- a $35M joint partnership of General Motors, Delphi Automotive Systems, and the National Highway Traffic Safety Administration (NHTSA).  In an exclusive interview, Ron Colgin, GM's Program Manager, and NHTSA's Project Manager Jack Ference give IVsource an update on system performance and future plans.

After over two years of extensive development and testing, General Motors and government engineers have certified a functional prototype for the Advanced Collision Avoidance System (ACAS) -- a milestone that gives the green light for Phase Two of the program.  ACAS is a joint partnership of GM, Delphi Automotive Systems, and the National Highway Traffic Safety Administration (NHTSA).  The Volpe National Transportation Systems Center in Cambridge, Mass., is serving as independent evaluator.

IVsource spoke to Dr. Ron Colgin of the Chassis and Vehicle Systems Department at the GM Technical Center, and Jack Ference, ACAS program manager within the NHTSA Office of Crash Avoidance Research, to discuss the latest status of the program.  The intent of ACAS is to evaluate the safety performance and driver interface of a combined Adaptive Cruise Control (ACC) and Forward Collision Warning (FCW) system implemented on 2002 Buick LeSabre automobiles.  Automotive supplier Delphi Systems is providing much of the technology used.

The aim of Phase One of ACAS -- now completed -- was to define the system, evaluate and select technologies, and demonstrate a fully functional prototype vehicle.

Radar and Vision Strut Their Stuff in Sensor-Fused System Approach

Within Phase One, Delphi provided evaluations of current technology, as well as new technology development.  Their production ACC system (currently on the market in Jaguar vehicles) was significantly enhanced to implement the FCW function, and vision-based techniques for sensing the road and lane configuration ahead were looked at.  Of several “entrants” in the vision technology evaluation, algorithms from the University of Pennsylvania were deemed superior for the task. However, after the chief researcher for the project at U. Penn left the university, Delphi took their approach and developed it further for the final design.  Delphi’s vision system looks forward about 80 m (260 ft) to perform lane discrimination, as well as “close-in” to assess lane placement and the situation around the car.

Radar data is also used to estimate curves ahead on the roadway by observing the general trajectory of preceding vehicles.

In addition to providing project management and systems engineering, GM’s major role is the challenging task of subsystems integration and sensor fusion.  They have integrated digital-map-based road geometry estimation (using current generation digital maps), developed target tracking and threat assessment software, and the driver-vehicle interface.

Using this sensor suite of vision, radar, and GPS, coupled with vehicle sensors (yaw rate, acceleration, etc.), the system performs scene tracking to interpret what is happening ahead of and around the vehicle, and to project the likely path of the vehicle.  Based on sensor-fused data, target selection is performed to define the closest in-path vehicle or object.  At that point 'threat assessment' kicks in, to determine if a condition is occurring for which the driver must be warned.

When ACC is functioning, brake activation is included for normal ACC deceleration.  When the braking authority is exceeded in approaching a slower vehicle or obstacle, the driver is warned.  When the ACC is not in use, the system operates as standard FCW, with the driver warned at the appropriate time.  Unlike many ACC systems now on the market,  the system “sees” stopped objects in the vehicle’s path – obviously an essential requirement for an FCW system.

The system can track vehicles around curves down to a 500 m (0.3 mile) radius -- a bit tighter than curves typically found on major highways.  Operating speed is 25 mph (40 kph) and above.

GM and NHTSA Algorithms to Run Side by Side

The overall software algorithm to interpret the many sensor inputs and sound an alert only when needed forms the heart of the system’s intelligence.  Researchers started with seven candidate algorithms and down-selected to two – one algorithm developed by GM and the other developed by NHTSA.  For evaluation purposes to support other research, the NHTSA algorithm will run in background mode and will be post-processed to determine its effectiveness as compared to the primary GM algorithm.  “We will publish both the algorithm and an analysis of its performance data once the test is completed,” says Ference, thereby enabling other developers to potentially refine their algorithms.

Phase One Completion Marked by Extensive Testing

During October of last year, 200 runs to conduct 30 different types of tests were performed to certify the system.  Testing included use of test tracks at GM’s Milford Proving Grounds as well as open highway driving.

A 100+ page report documents the comprehensive testing performed.  Testing included:

• Stopped object in vehicle path

• Hard braking of preceding vehicle

• Following a motorcycle

• Following a motorcycle which is following a truck (a very challenging situation for radar systems)

• Tracking vehicles on curves

• Responding to sudden “cut-ins” in which a vehicle in the adjacent lane moves into the small gap between the subject vehicle and the target vehicle

• Assessing the frequency of nuisance alerts

In addition to testing at Milford, 188 miles (300+ km) of testing was performed on public roadways.  A combination of highways, arterials, and rural roads was selected to reflect the ratio of these types of roads in the US highway system.  Weather conditions included both rainy and clear weather.

As with any testing phase, several problems were uncovered.  Researchers identified a system bug that occurred when the vehicle was going around a curve – detection of threats was very late.  (It turned out that the target acquisition subroutine in the software was creating data overflows in these conditions, which caused a data buffer to reset and require re-detection.)  Also, some initial poor test results were traced to a radar beam pattern that was found to be miscalibrated to be too narrow, creating a limited field-of-view.

The Bridge Problem -- Nailed;  Curves Still Tricky

After the aforementioned glitches were fixed, testing proceeded with good results.  For the open road testing, only two nuisance alerts per 100 miles on roadside objects were noted, according to Colgin.  Of 148 overhead objects (typically bridges), an astounding zero nuisance alarms occurred from the system mistakenly seeing the bridge as a forward obstacle -- a major challenge for designers of early ACC systems.

However, curves on rural roads continue to prove to be the most difficult situation.  “The system would tend to create false alarms coming around a curve where a mailbox was on the roadside ahead,” said Ference.  But overall, rejection of roadside elements works pretty well.  Guardrails are fairly consistently rejected as obstacles, and Ference tells of a particularly cluttered construction zone in which the system successfully recognized and rejected items such as Jersey barriers and nearby construction equipment.

In separate test sessions, false alarms occurred on 1 out of 22 curves and 3 out of 12 curves.  In multilane settings where vehicles were passing in the adjacent lane, 32 encounters occurred which produced six false alerts over eight hours. 

“This level of performance is fine for initial consumer use,” Ference believes.  He drove the prototype vehicle for four hours in Michigan.  “It felt comfortable and operated the way I thought it should.  There were not many false alarms,” he said.

Independent Evaluators Played Key Role

The government funded the involvement of independent test witnesses from the Johns Hopkins Applied Physics Laboratory.  “APL did a great job,” says Ference. “They really got deeply involved in the technical aspects, which enabled them to find problems in the test protocol and some bugs in the algorithm software.  Thanks to them, we knew what was going on.  They were key to the process.”  Over the last two years, APL has received $400K in funding from NHTSA for supporting this type of research.

In fact, Ference had nothing but praise for the entire industry team.  “To GM’s credit,” he continued, “they performed the testing very well and in minimum time.  The collaboration amongst all the players was very strong.

“Phase One really came together well into a very polished system.  We learned what is going to work well and what we have to continue to work on.”  He is particularly pleased to note that the project became a “truly cooperative venture” between Volpe, APL, GM, Delphi, and NHTSA.  “The team was close-knit, they had mutual respect, and data flowed freely between the players.  Overall it has been a good process.”

Some Technical Tasks Cut Out for the Next Phase

Although the prototype performance has been officially accepted, tweaking is still ongoing.  The ACC algorithms are being refined to smooth out braking.  Further development of the data fusion approach is underway, and project managers say they have more to learn about the vision system.  For instance, lighting conditions sometimes present problems, and the full desired capability has yet to be achieved.  The alert tone also continues to be revised and evaluated (the current tone is said to sound like an electronic pager and may not be “urgent” enough).  Researchers are trying to find the balance between creating urgency and not overly alarming/irritating the driver when a false alarm occurs.

Reports Being Prepared for Publication

As the project progressed in 2001, quite a few detailed reports were generated and delivered to NHTSA.  At this time, however, NHTSA is only planning to publish two reports.  The project interim report is expected to be available by the end of March, and the Warning Cue Implementation Summary Report, focusing on the driver-vehicle interface, is set to be released a few weeks sooner.  Ference says that these will be available on a compact disc and he is happy to mail a CD to anyone who asks.   (See contact information below).

Ramping Up for Phase Two Vehicle Builds

Dr. Colgin tells us that parts are now on order to begin the vehicle builds – by March 2003, 13 vehicles will be equipped with a production-quality ACAS system.  The first pilot vehicle is being built during March through May 2002.  Once the first four vehicles are complete, researchers at the University of Michigan Transportation Research Institute (UMTRI) will perform evaluation runs with lay subjects to evaluate the effectiveness of the driver-vehicle interface and driver’s overall reactions to the system – areas such as “look and feel” and the smoothness of the ACC will be of particular interest.  These runs will be conducted with a researcher in the vehicle as an observer.

Once the field operational test begins in December, drivers will use the cars in their normal daily business, with no researchers or obtrusive evaluation equipment involved – the vehicles are meant to look and drive as if purchased from the showroom.  The operational testing will last for ten months.

Cost Sharing Variable as Project Progresses

Costs for Phase One came in just under $20M total, with this developmental phase split roughly 50/50 between industry and government.  The government will cover roughly 80% of costs of the operational testing (Phase Two).  Of the total $35M project, GM is contributing $7M, Delphi  $7M, and USDOT  $21M.

So ... When Will These Systems Hit the Market?

GM executives indicate that Adaptive Cruise Control can be expected to be introduced this year on their Cadillac model.  Offerings of a full-up ACC/FCW system, such as that implemented in the ACAS program, is still a few years away.  However,  sources tell IVsource that market introduction before the end of the ACAS program in 2004 is within the realm of possibility.

Meanwhile, the curious can hear a presentation on the project by Dr. Colgin at the ITS America Annual Meeting in May, and it is likely that GM will bring the Phase One prototype car to the National Intelligent Vehicle Initiative meeting later that month.  They are also considering demonstrating ACAS cars at the ITS World Congress in Chicago this fall.

[Top]

For More Information ...

... contact Dr. Ronald Colgin at ronald.c.colgin@gm.com or Jack Ference at jference@nhtsa.dot.gov.

[Top]