- Jun 9, 2006
Head of Stanford DARPA Challenge project speaks about the future of autonomous vehicles
We had the pleasure of catching Dr. Sebastian Thrun's keynote presentation at the 2006 Sensors Expo on Tuesday, where he spoke to the audience about his experience leading Stanford's DARPA Challenge entry to victory last year and how the project's success may affect the way we drive in the not-so-distant future.
The technical content of "Stanley" is fascinating, of course. Due to the relative inaccuracy of GPS (2 meters of error doesn't cut it on a 2.5-meter-wide mountain pass), a variety of sensors were used for environmental recognition, allowing the vehicle to discern "good" terrain from non-drivable areas. A array of five lasers scan the area in front of the vehicle for obstacles, and their downward trajectory combines with the forward movement of the vehicle to allow for the return of 3D data. Unfortunately, the 20 meter range of the lasers and the 70 Hz scan rate limits the maximum practical speed of the vehicle to 35 MPH or so when relying solely on this navigational device.
To allow the faster travel speeds thought necessary to secure a victory, an optical camera was implemented to learn from the laser system how drivable terrain appears, eventually allowing the vehicle to see much further down the intended path and thus facilitating higher speeds. Doing so isn't as easy as it may seem - color can't be used due to the wide variety in coloration of drivable surfaces (roads may be brown, black, red, tan, or any number of other shades), and texture detection is nearly worthless as Thrun noted that the sky is smoothly textured but yet not a practical path. Radar is also used to provide ranging information.
[Click through for much more on this vehicle and the future of (not) driving...]
Eventually, the vehicle was able to autonomously travel fast enough to scare its human passengers, and so shock sensors were employed to slow the vehicle after hitting something too fast. It was admitted that this reactive system is imperfect compared to the anticipation provided by a human driver, but for now it's considered to be a workable solution.
Thrun stated that the team's success depended heavily on collecting physical data from actual driving to allow refinement of each system. Amazingly enough, the vehicle's autonomous navigation system provides all-weather capability, as proven by a video that shows the vehicle driving at high speeds across the desert in a torrential downpour (the vehicle's windshield wipers weren't even keeping up).
Thrun noted that each of the teams considering this to be much more of a contest versus nature and physics than against each other, and was careful to point out that the 11-minute gap between the finishing time of Stanford and Carnegie Mellon's second-place entry is not significant in the grand scheme of things. Instead, it's more important to focus on the fact that the competition evolved from a complete failure to one where five vehicles finished in only a year. This shows the rapid pace of advancement in autonomous navigation.
Looking past the competition and towards the future of transportation, Dr. Thrun threw out some sobering statistics. Traffic fatalities are the leading cause of death for those aged three to 33 years, and 1500 people die every year from collisions with trees - a slow-moving target if there ever was one. While gridlock remains an ever-increasing concern around the world, the average stop-and-go situation still involves roadways that are 92% free of cars; that is, due to a human's need to maintain some space to provide for sufficient reaction time, we are using less than 10 percent of a road's capacity. Essentially, people are poor drivers, and autonomous navigation will offer the ability to offload certain tasks from the wetware behind the steering wheel.
It was acknowledged that this would not happen overnight; rather, driving aids such as distance-following cruise control (including the upcoming stop-and-go versions that are useful in urban environments) and lane-departure warning systems would be the first step. Eventually, Thrun thinks that such technology would be able to improve urban environments by allowing the remote location of parking lots (your car might drop you off at the office and then park itself miles away), and could also improve the mobility of senior citizens as life expectancies continue to increase. Thrun injected a personal anecdote at this point, telling about how his family recently had to take the keys away from his elderly father. Such a collapse of one's social network often leads to a senior's demise, and as such autonomous navigation could lead to longer lives.
By implementing autonomous navigation in individual vehicles instead of trying to build "smart highways" (those with built-in navigational devices, such as magnetic strips), Thrun feels that the benefits of such technology can be brought to the average driver much more quickly and at a lower overall cost. It was suggested that perhaps special lanes could be dedicated to autonomous vehicles as a starting point, perhaps much as car-pool/HEV lanes are used today.
During the Q&A session, it was asked why an American university didn't use a domestic vehicle. Thrun stated that Ford offered some level of support but generally was not interested in being associated with a military project or the aura of failure that surrounding the inaugural DARPA Challenge, and GM showed no interest in the competition. It was said that, generally speaking, there is far more interest in autonomous navigation coming from Europe and Japan than there is from the US.
Certainly, after watching this presentation and gaining a far better appreciation for what has been accomplished and potential lies aead for autonomous navigation, we await the upcoming Urban Challenge. For this event, Stanford will be fielding a VW Passat.