This simulator wants to solve a big problem with self-driving cars: you
Self-driving cars are edging ever closer to becoming smart, reliable motors people can actually buy from their local dealerships. Google's crafted its first cheery prototype and Audi's driverless RS7 will happily whip around a test track at 150MPH. But while autonomous-vehicle technology is maturing, engineers and researchers still have plenty of challenges ahead. Many of them revolve around human interaction -- when driverless cars finally enter the public domain, they're likely to come with standard controls, including pedals and a steering wheel. Despite their ability to cruise around independently, these vehicles will require a manual override just in case the driver needs to retake control. But how exactly will that human-machine changeover play out inside the vehicle?
In the sci-fi movie I, Robot, Will Smith's character takes control of an autonomous car when it's attacked by androids. A quick button-press summons a steering wheel from beneath the dashboard, allowing the detective to take evasive action. The process looks natural and seamless on the silver screen; right now the reality is anything but.
The challenge is creating a system that's both safe and comfortable for the driver. The car could release control immediately, for example, or check first to ensure they're alert and the road conditions are safe. In the latter scenario, the system could count the driver down as it analyzes the situation, or offer basic assistance while they adjust and retake control.
The Transport Research Laboratory (TRL) is researching this problem as part of its new, government-funded driverless car project. With a simulator called DigiCar, which attempts to recreate the driving experience and record passenger behavior, it can analyze how people react to autonomous systems when they have to retake control on the road.
Located deep in the heart of Wokingham, the test room is oppressive. TRL's chosen vehicle, a small Honda Civic, is bolted to the floor with tall, angled screens facing its front and derriere. Projectors stream a virtual French Riviera -- this will be our test track.
While the car is familiar, its operations are anything but. When you turn the key, the engine doesn't spring to life, so finding the biting point and rolling away in first gear seems almost impossible at first. TRL has done its best to inject some realism, though. The speedometer works properly and extra speakers inside the car attempt to reproduce the noise and vibration of the engine. It's even kitted out with electric motors that make the car pitch and roll where appropriate. Although everything is in sync, it was never enough to fool my senses. Sure, I was changing gears and (occasionally) checking my wing mirrors, but there were no real dangers on the road. TRL's virtual world is primitive too, with low-resolution objects and textures making it more Sega Rally Championship than Driveclub.
But that's almost beside the point. DigiCar is designed to test how people react to self-driving cars and understand the psychology behind their actions, not win awards for photorealistic trees. Inside the Civic, my instructor could switch the car among autonomous, manual and two assistive modes using a tablet. Switching modes is near-instantaneous, although my guide was nice enough to give me a quick heads-up before removing and returning control. With limited feedback from the vehicle, the transitions are pretty jarring, and TRL was tracking all of my clumsy reactions. The simulator records steering angles, gear choices, pedal depression and even engine RPM. In turn, eye-tracking cameras and heart rate monitors record driver information, explaining to TRL how they react in different situations. These measurements are combined with simulation data, such as the car's speed, position and distance from other objects, which the team can supplement with traditional driver interviews once the test is complete.
DigiCar isn't a perfect representation of the real world, but that doesn't mean its findings won't aid autonomous vehicle development. "The information will be useful to vehicle manufacturers in developing and refining their systems," Nick Reed, academy director for TRL said. "It's also useful for regulators in determining how they need to regulate these systems and their implementation on real roads." In addition, Reed says the data benefits insurance firms looking at the impact of these systems on driver risk and liabilities.
Even before the new government funding, DigiCar was producing some interesting results. At one point, TRL was looking at how automation could manage "platoons" of trucks driving single-file down a motorway. The drafting technique improves fuel efficiency, so the team linked some vehicles together to automatically manage their gaps and speed. When drivers encountered the trucks inside the simulation, they would reflexively reduce the distance between their own car and others around them. The practice was contagious and TRL noticed that test subjects would maintain these smaller gaps in their regular driving. While not directly related to driver-machine interactions inside autonomous vehicles, it was an unexpected finding that could still be useful to regulators and manufacturers.
How DigiCar research can be used to adapt self-driving car systems is another matter entirely. Nevertheless, any interactions that would normally be too risky to observe on public roads can now be tested within the relative safety of the simulator. It's a handy facility to have in the UK, and one that goes beyond mere hardware to look at how we actually feel about giving control to, and taking it back from, a robot chauffeur.
The DigiCar is just one of three driverless car projects we looked at this week. Check out part two and part three in our miniseries.
