Safety, pah! We have robot cars powersliding, say US boffins – The Register
Last month, Google, Ford and Uber started a lobbying group called the Self-Driving Coalition for Safer Streets to promote autonomous vehicles on safety grounds, but boffins in Georgia are having far more fun with their hardware.
While Google spends time pootling around Mountain View at 25mph, researchers at Georgia Tech, the state famous for Bo, Luke, and Daisy’s driving habits and their General Lee Dodge Charger, have been testing robot cars powersliding around a dirt track at the equivalent of 90mph using scaled-down cars.
“It’s the ability of the car to perform what are called aggressive maneuvers,” said Jim Rehg, professor at the university’s school of interactive computing.
“So doing turns, for example, the car is literally sliding around the turn.”
The metre-long electric rally cars house a quad-core i7 processor, Nvidia GTX 750 Ti GPU, and 32GB of RAM; GPS; and twin cameras in a crash-hardened shell. That, and the software algorithms developed by Georgia Tech, enables the car to game out over 2,000 projections of where it’ll head in the next 2.5 seconds.
The system (whose specs have been published for general use) performs this calculation 60 times a second and puts the power down as soon as possible for maximum velocity. It takes a few circuits with a human controller to learn the course, and then the computer takes over and starts putting the pedal to the metal.
In a paper at the International Conference on Robotics and Automation, the team detailed how the cars – cosponsored by the US Army Research Office – were able to learn how to handle different tracks via trial and error. There were a few crashes, when the software couldn’t handle the road surface, but the research showed how autonomous cars could react to changing circumstances on the road.
“Aggressive driving in a robotic vehicle – maneuvering at the edge – is a unique control problem involving a highly complex system,” said team leader Evangelos Theodorou.
“However, by merging statistical physics with control theory, and utilizing leading-edge computation, we can create a new perspective, a new framework, for control of autonomous systems.”
While the research may help keep passengers in autonomous cars safe in difficult driving conditions, it’ll be of little help in persuading Congress to allow self-driving cars on US roads.
Google, Apple, Ford, and Tesla are all lobbying on the grounds of robot cars being safer than flashy drivers, and the Chocolate Factory’s tests seem to bear this out. So the prospect of digital Top Gear presenters bombing around the roads in chipped robo-cars might not be seen as helpful to the cause. ®