The future of autonomous vehicles remains clouded by safety concerns, prompting manufacturers to offer only limited options for a car to take over, such as in highway traffic jams or parking lots.
But Chris Gerdes think the answer to designing a safer autonomous car might be found in pushing them to go faster - a lot faster.
The Stanford University professor and senior fellow at the Precourt Institute for Energy has for the past few years studied how professional racing drivers make their way around the track at speeds upwards of 150 miles per hour. He then inputs that data into his automated Audi TTS known as Shelley.
The testing goes on at several tracks including the Bonneville Salt Flats and Thunderhill Raceway Park in California, where Gerdes examines such things as how David Vodden, an amateur driver and CEO of Thunderhill, or Gunnar Jeanette, a current driver who has competed in the World Endurance Championship, brake, steer and shift gears.
Gerdes uses cameras inside and outside the vehicle as his drivers take on the track to see how they handle turns and straightaways, when they cut close to the inside, when they go wide, and when and how they use the steering wheel, the pedals and the gear shifter. Gerdes also attaches electrodes to their heads to track their vital signs and brain activity as they make their driving decisions.
"It's easy to see what they actually do with their hands and how the car moves out on the track," Gerdes said. "We have been taking brain wave data and trying to correlate that with specific vehicle actions. We can see some general trends."
He then takes all the data gathered from the cameras and electrodes and feeds it into Shelley's program, teaching the car how skilled human drivers would behave in certain situations.
"They try to use all the friction capability of the tire and the road to be fast," Gerdes told CBS News. "What we really want to do is have that same capability to use all friction between the tire and road to be safe. If the laws of the physics say the accident can be avoided, we'd like the car to have capability to avoid it, whether that is fully automated car or a car that is working together with a human driver."
Ever since Google first started testing its self-driving car, the hype was that vehicles operated by computers would eliminate the human error that led to 5.7 million motor vehicle accidents in 2013. More than 32,000 of those crashes were fatal, according to the National Highway Traffic Safety Administration. Much of the blame came down to the drivers being drowsy or intoxicated. The consulting firm KPMG found that 93 percent of accidents are caused by human error.
So it might sound counterintuitive that to make a safer automated car, Gerdes and other engineers think vehicles need act a bit more human. They are finding that one of the limits of automated cars is that they lack the flexibility and judgment to respond to the array of hazards humans confront every day on the road, from snowstorms to construction sites to the errant pedestrian crossing our paths.
Some of those limitations have been on display out at Thunderhill, where Gerdes routinely sends Shelley out on the winding, 3-mile road course.
"When we put car out more robotically and we say OK follow this path, it tends to think all points on the path as equally important," Gerdes said. "But human driver will say there is a point where I'm putting my wheel up against the edge, which is extremely important. If I allow a little bit of error in middle of track, that is not such a big deal if I can get extra speed, extra stability. That is really good trade off."
Vodden, an amateur racer for nearly four decades, said, "There are a lot of techniques that human brain learns and perfects to go faster around the race track and beat the other drivers."
In the early years, he and other drivers managed far better times around the track than Shelley because they could make adjustments to cut their speed. But Shelley is proving to be a fast learner. As fresh data gleaned from the drivers is imputed into its system, the Audi has begun charting a path around the track that is almost as fast as some of the race car drivers - a response that could be especially beneficial, say, on winter roads in the decades ahead.
"Imagine the race car comes into corner and they are trying to push the car as fast as possible through that corner without it spinning or without going off the track," Gerdes said.
"Now an automated car may come into corner and discover it's very icy. They are in the same situation of trying not to spin out and trying to get through the corner. The car has been pushed to the limits not because someone is racing on the track but simply because you hit a patch of ice."
Gregory Fitch, a research scientist who leads the User Experience Group for the Center for Automated Vehicle Systems at the Virginia Tech Transportation Institute, said it made sense to study the reaction of race car drivers.
"If you were to develop an understanding of when they experience stress in certain conditions, you can pretty much guarantee that the normal driver going to experience stress in those conditions," said Fitch, who just finished an extensive study of driver engagement with the Google self-driving car at its track. "It's kind of a way of identifying edge cases for people's reaction to how a vehicle might drive."
And by studying a range of driving styles including professionals, Fitch said it will help engineers move away from a one-size-fits all model.
"Flexibility is part of the key to driver acceptance," he said said. "I think user acceptance will dependent on whether or not the driver approves of how the car is driving. I think it makes sense to look into adjustability. I think that makes complete sense."
That is where the brain waves could help.
The challenge now is making sense of the data. "You don't really know what caused certain brain activity," he said. "Was it directly related to what is on the track or was it some stray signal? Was it them thinking about what they had for dinner the night before?"
Gertes said they are trying to better understand how to interpret the brain wave patterns they're observing and possibly even use that information to help drivers in the future.
"One thing that would be really interesting is to be able to measure somebody's skill with handling a car, to understand where a car should be supporting the driver," Gerdes said. "Can we actually see in brain patterns certain driving skills and know whether people acquired them or not? Can we have the car actively train the driver?"