“Move fast and break things,” the phrase popularized by Facebook CEO Mark Zuckerberg, serves as an unofficial mantra for many people and companies working in the tech industry. But are they really words to live by when it comes to self-driving cars? If you ask the Georgia Institute of Technology researchers behind a project called the AutoRally Robot, they may well answer in the affirmative.
The AutoRally initiative, which is the kind of project that the cast of The Fast and the Furious movies might suggest as a Ph.D. thesis, pushes self-driving cars to their limit by getting scaled-down autonomous vehicles to drive really, really fast and aggressively on dirt roads — similar to a rally race.
“The sensors on AutoRally are a subset of what you typically find on a full-sized self-driving vehicle: GPS, IMU, wheel speeds, and a forward-facing camera pair,” Brian Goldfain, a researcher working on the project, told Digital Trends. “We are experimenting with lidar sensors, but it has proved difficult to protect it from the frequent rollovers encountered during testing, [while mounting] the sensor high enough for an unobstructed view of the environment.”
The perception algorithms developed by the team use only camera inputs, which are cheaper, more robust, and can “provide much richer information” than the typical lidar, radar, and ultrasonic sensors found on self-driving cars. Driving based purely on camera images is much tougher for an autonomous driving system to wrap its head around, particularly when the whole thing is happening at dangerously high speeds.
In the cars’ efforts to rack up faster and faster lap times, the vehicles even perform maneuvers such as drifting, or controlled slides around corners, which wouldn’t be acceptable in many driving environments. That’s why it’s a good thing that the AutoRally is taking place on a specially constructed dirt track in Georgia, far away from pedestrians.
Ultimately, despite the reckless driving of the vehicles, the team hopes that its work will bring safe self-driving vehicles to public roads sooner. “The [eventual] goal of the AutoRally project is to enable the next generation of algorithms for self-driving vehicles.” Goldfain said.