We keep being told the future of transport is autonomous; our vehicles are going to end up driving themselves. But it may be of comfort to driving enthusiasts that computer control has a long way to go, and there are still things that humans do best behind the wheel.
“In the short term at least, there are strengths that humans still bring to the table, and we don’t want to rule them out altogether,” says Stephen Erlien, of Stanford University, in Silicon Valley.
Still, sometimes even the best drivers can get distracted. Could next-generation cars step in to take over the wheel?
To find out more, I went to Stanford’s Center for Automotive Research (Car) to meet neuroscientist Lene Harbott. The plan was to hook me up to an EEG machine that measures electrical activity in my brain, while Erlien took me for a spin in Stanford’s experimental vehicle, the X1. This car, which looks a bit like a moon buggy, was built by students as a testbed for new technology, and is currently being used for automatic hazard avoidance tests (see video above).
As I drive the car, my brain waves will be recorded, to help figure out how hard I’m concentrating. Harbott calls this ‘cognitive load’. “Rather than taking data from vehicles themselves, I take data from the human drivers,” says Harbott.
Before cars become fully autonomous, there are going to be aspects of shared control with the human driver. It is already happening with the types of active safety controls we are seeing on modern high-end car. So what Harbott is trying to understand is how hard drivers have to work mentally under various driving conditions, and also whether these ‘driver assistance’ devices (such as adaptive cruise control which can slow and even stop a car if the vehicle in front stops) actually help with that mental work load. There is a possibility that a badly designed system can make it even harder for a driver to concentrate if it presents them with too many warnings.
Recording brainwave data from drivers is a challenging task. Harbott picks up five narrow cables, coloured green, red, blue, yellow, and orange, each with a flat metal tip like a small coin. The first one is glued onto the bony area behind my ear, to give a reference reading, and then Harbott sticks the others in a line down the centre of my skull. The end result looks like multi-coloured hair extensions.
As Harbott works I ask her how a neuroscientist came to be working in the automotive industry. “I have a love of motor sports inherited from my Dad, particularly vintage motor sports,” she tells me. “We grew up not far from Silverstone [in the UK]. So we used to go and watch the vintage racing all the time.” Stanford was already running programmes to collect as much data as possible from cars, and she suggested collecting data from humans would be useful too.
Once I’m wired up, Erlien leads me to the X1. I clamber into the low vehicle, and settle into the racing seat, buckling up the four-point safety harness. The vehicle is a mass of cables and components, and Erlien, sitting in a passenger seat with a laptop, reads their resulting data. It also allows him to adjust parameters on the fly. We set off around the campus at Stanford, dodging cyclists and other traffic. I am already having to concentrate quite hard. Then Erlien plays with the settings, and suddenly the car has four-wheel steering and handles totally differently.
Tests of the X1 at Thunderhill Raceway in Northern California subjected the drivers to even more extreme changes – I get off lightly as a newbie. They were suddenly thrown into a slide by scientists who programmed the rear wheels to move. The driver would have to apply a quick oversteer correction to keep from spinning out. It’s an emulation of hitting a patch of ice, or oil. Data is collected from the car, showing the steering wheel angle (ie: which direction the driver is trying to steer) and the pedal positions.
When this gets interesting is when this data is plotted into a graph of jagged peaks and troughs, and compared to the brain-wave readings. The peaks and troughs align, showing a neuroscientist like Harbott how the cognitive load spikes when the driver encounters something unexpected.
I ask Harbott about notches that I can see in the graph, showing wheel angle. She explains that they align with when the back wheels have slid out - which is when a driver would suddenly have to concentrate, and apply steering correction to avoid a skid. Comparing that to the bottom graph of brain waves shows similar changes. "Here you see different energies in the different frequency bands that we talked about," she says. "So as the lower frequency bands are increased, that’s a sign that the driver’s working harder mentally than at other times." This work is ongoing and the results require an expert eye, but they do seem to show an effective way to measure how much the driver is having to concentrate. Those spikes happened when I encountered one of the many cyclists on Stanford's campus, or when Erlien changed the driving characteristics of the car.
“One of the things that’s interesting with these experiments is that the more times the driver experiences this, the less difficult it becomes cognitively,” says Harbott. “So, you train yourself to do it.” So this suggests that if drivers train in icy conditions, for instance, they will be better able to cope in a real setting.
Crucially, these observations lead to the idea that safety systems could be developed to directly assist with those tasks.
“In an emergency situation most of your cognitive load is taken up dealing with step one,” says Harbott. “But things tend to come at you thick and fast in an accident, and you may have used up all of your cognitive load, as it were, dealing with the first thing – then things just domino. So if the car takes some of that work away from you, you have more cognitive load to spare if you’re in an accident.”
So what the team at Stanford is learning from humans will eventually lead, perhaps, to more intelligent computers that lead us away from the steering wheel.