Chris Gerdes, professor of mechanical engineering, is the director of Stanford’s Center for Automotive Research. The Daily sat down with Gerdes to discuss his recent work on mapping the brain activity of racecar drivers to improve self-driving cars.
The Stanford Daily (TSD): How did you come up with the idea of trying to monitor the physiological activity of racecar drivers in order to improve autonomous cars?
Chris Gerdes (CG): The instigation for that whole project was a result of work with Cliff Nass. When Cliff and Michael Shanks and I started the Revs program at Stanford, we had the idea of looking at the car from different perspectives. And so we had the idea of working with some of these vintage racecars. My part of that was going to be to instrument the car, and Cliff’s part of that was going to be to instrument the driver. So we started to work on this cooperatively. Then when Cliff passed away, we continued in my lab.
TSD: How did you deal with ambiguity in the physiological data that you’ve found – determining the source of the brain activity?
CG: Well, that is what we’re still working on, as a matter of fact. And the work is continuing together with some collaborators in the medical school. The fundamental problem that we have when looking at brain waves is that we can only measure general electrical activity in the brain from electrodes that we place on the scalp. The description I once heard was that it’s is like trying to stand outside a football stadium and figure out what’s going on in the game by the crowd noise. There are certain things that you get out, but you don’t get all of the details. We’re able to see patterns in particular frequency response patterns — relative proportions of alpha waves or theta waves in the brain – at different times when the driver’s on the track. The difficulty is, as you said, with the ambiguity. How do we know if that particular pattern was a result of a driver having to focus on taking the turn or the driver thinking about when he could pass the driver ahead or the driver thinking about what he had for dinner last night? That’s really the challenge.
TSD: How will you be able to take this research you’ve done and make it applicable to highways or other simple environments that a self-driving car would have to navigate?
CG: Although we can hit very high speeds on the straightaways of the racetrack, most of the time when we’re going around the turns, we’re somewhere close to 50, 60, 70 miles an hour. So in fact, the car is actually maneuvering at what would be highway speeds. If you had to do an emergency lane change on the highway, which is something we’ve programmed Shelley, our self-driving car, to be able to do, then these techniques apply directly. If I suddenly have a car swerve in front of me, and I need to change lanes, I want to use all the friction available between the tire and the road to do that safely. Shelley has that capability, and it’s a direct translation from racing. The other thing is that if you come into a curve too quickly, and the road is icy, it’s the same physics that describes the motion of the car there as describes the car on the racetrack. It’s just that friction is greatly reduced. So if we can figure out how to keep the car stable and on the road in racing, we’ve also solved the problem of how we keep the car stable and on the road when it hits an icy surface.
TSD: How do you hope to expand this project?
CG: One of the big things that this research has led us to think about is the comparison between what human drivers are doing and what the car is doing out on the track. Shelley is designed to follow a desired path very closely and try to travel that path at the greatest possible speed. Human drivers have a lot more flexibility in their approach; they have an idea of the path that they want to follow, but they’re very happy to bend it as they see opportunities or as they feel little nuances in the vehicle dynamics. Shelley’s capable of changing paths, but this more nuanced approach of constantly changing based upon conditions is something I think we can still learn from racecar drivers.
TSD: Considering the fact that human drivers often have to react to conditions in a way that may break the law, how will you balance the efficiency of a car and its ability to deal with context with its ability to follow the law?
CG: That’s a great question. If we think about humans not really being beholden to a particular path, one example that comes to mind is if I’m driving down a two lane road I have a double yellow line that says I’m not supposed to cross it. And I have a bicycle on the right side of the road. I think many of us in that situation will cross the double yellow line in order to provide some extra measure of safety for the bicyclist. But that’s actually a violation of the California state Vehicle Code. We have violated the law in this case, but we’ve done it with the best of intentions. Our desire for mobility, to keep moving down the street, prevents us from just following the bicyclist at a safe distance. Our desire for safety suggests that we should give more room to the cyclist. And both of these desires really overcome our desire to follow the law in that case. I think this is going to be an increasingly important issue for autonomous vehicles.
Contact Skylar Cohen at skylarc ‘at’ stanford.edu.