Autonomous RC Truck Drifting on Purpose Gives Us Hope for the Future

Do you remember how people used to snub electric cars for taking the fun out of driving, and then Tesla came along with the Roadster and proved everybody that was all just nonsense? Well, a similar revolution needs to happen with the autonomous technology as well, but how could that be?
AutoRally drifting 1 photo
Photo: Screenshot from YouTube
How could self-driving cars make driving fun, when their entire existence is based on relieving people of the task of controlling a car themselves? That makes absolutely no sense, right? Well, yes. But also, no. Yes, autonomous cars do want to take over the wheel from us, but no, that doesn't necessarily mean it's got to be boring. And we're not talking about the fact that it frees you to watch a comedy show or something. No. It would appear somebody at Georgia Tech figured out that knowing how to drift is just as important for the future robotic driver as going steady in a bend.

That's not something you hear people talking about every day, but at a closer inspection, it's absolutely relevant. During the many billion of miles that autonomous cars will cover once they break into the mainstream, some of them are bound to lose traction now and then. And some of these cars will do so while turning. What will the AI do then? How will it know how to handle the situation if it was never taught anything like this?

The researchers used a heavily modified RC truck as their study platform, which they equipped with GPS, IMU, wheel encoders, two fast video cameras, and a quad-core i7 computer with an Nvidia GTX 750ti GPU and 32 gigabytes of memory. The vehicle is about one meter long (three feet), weighs 21 kilograms (about 46 pounds), and is capable of reaching speeds of up to 100 km/h (62 mph). They call it the "AutoRally."

The AutoRally was set loose on an oval dirt track where it was asked to maintain a speed of eight meters per second as constant as possible without crashing. Everything you see in the video below is the CPU doing the driving, even though there was a brief learning process first where a human drove the AutoRally in remote-control mode.

After that, it was all down to Georgia Tech's method of combining the two algorithms for steering and throttle by having the CPU project 2,560 different possible trajectories at any given moment. These trajectories are refreshed 60 times per second, and they represent the next 2.5 seconds of the vehicle's motion, so there's plenty of work to do for that monstrous processing power.

Spectrum says that the Georgia Tech researchers have made everything about the AutoRally public, posting the software it uses on Github in hope that everybody interested in adding a new side to their autonomous vehicle would pick it up and use it. The video below is quite long, but it does explain everything rather well, so it's worth giving it six minutes of your time. If you're in a hurry, you can skip to where the autonomous AutoRally makes the most of the low-friction surface by powersliding all over the place by fast forwarding to the two-minute mark. There are also some crashes after the fifth minute, if you're that kind of a guy.

If you liked the article, please follow us:  Google News icon Google News Youtube Instagram
About the author: Vlad Mitrache
Vlad Mitrache profile photo

"Boy meets car, boy loves car, boy gets journalism degree and starts job writing and editing at a car magazine" - 5/5. (Vlad Mitrache if he was a movie)
Full profile


Would you like AUTOEVOLUTION to send you notifications?

You will only receive our top stories