Add to favourites
News Local and Global in your language
20th of July 2018


Crash-Happy RC Cars Make Self-Driving Tech Smarter

The road is a messy, chaotic place. Unpredictable drivers. Ice. Gravel tracks. Kids dashing into the road between parked cars. Hard to spot, hydroplane-inducing puddles. Surviving it all means knowing how to slam the pedals or swing the steering wheel without losing control. It's enough to make any robocar quiver. Moreover, the merciful rarity of these situations also makes them harder to prepare for. No practice, no perfection.

A team of self-driving engineers at Georgia Tech believes practical experience is an essential complement to simulation training. But real, prototype, autonomous vehicles are expensive. It can take an investment of $1 million to build up the sensors, computers, and controls an AV needs. They’re machines to be babied, driven gently, and definitely not taken to aggressive extremes. If they crash, roll, or skid off the road, they could put people in danger, and budgets in the red.

While that might not bother a giant like Audi, which has sent robots around race tracks and up Pikes Peak, it's a problem for anyone who doesn't have cash to burn.

“As researchers we don’t have access, time, money, or space, to work with full-sized vehicles,” says Brian Goldfain, a PhD student in robotics at Georgia Tech. That didn't stop him and his colleagues from doing their work. “We decided to create a fleet of scaled, self-driving vehicles.” These near-indestructible minicars can be used in any situation where researchers want to test self-drivers, but don't have the resources to run a regular vehicle—especially one that they want to push to its limits.

“Most of our experimental results focus on the task of off-road racing,” says Goldfain. That includes dirt tracks, where skids and rollovers happen on a regular basis during tests. But the team has made instructions on how to build the vehicles available to anyone online, so other researchers can experiment in any extreme situation they can think up. They've named it AutoRally: an open platform for aggressive autonomous driving.

The result is a 1:5 scale autonomous vehicle that looks like a miniature, battle-hardened pickup truck. The design is around three feet long, weighs 45 pounds, and can reach 60 mph—a harrowing speed for such a small machine. “It’s terrifyingly fast," according to Goldfain. "Our algorithms have never driven it quite that fast."

Brian Goldfain

Even if they did, the vehicle would likely be OK. It can spin, roll, and slide around in the dirt without getting damaged. Learning how an autonomous vehicle reacts in that situation could help develop one that can swerve away from that kid running into the street without losing control and hurting somebody else.

Instead of expensive lidar sensors, the little robots use a pair of forward-facing, well-protected cameras and stubby, roof-mounted antennas that allow researchers to hook up remote-control systems or collect satellite GPS signals, if that's called for by their experiments. That may not be representative of the sensors of a full-sized AV, but Goldfain says work done in miniature does scale up. Many of the algorithms researchers need to test in path planning are sensor-agnostic. They need some data about the world, but that could come from radar, lidar, vehicle-to-vehicle communications, or a camera. The basic neural net that is being trained will have the same structure.

Using the AutoRally platform already has allowed researchers to publish insights into autonomous driving with low-cost sensors, and low-quality input data. The findings are being presented at the Robotics: Science and Systems conference, on now in Pittsburgh.

By publishing AutoRally’s design and specifications, Goldfain hopes that other researchers can recreate his practical, hands-on approach. His team at Georgia Tech has built six of the robocars so far, and plans to get them all out on the track together soon. In the future, they could invite other researchers to run with them, or challenge their software. Think Battle Bots, but where the objective is to not collide with your competitors. It's a goal everyone will benefit from when autonomous vehicles are finally ready to roll along the roads.

More Great WIRED StoriesInside the crypto world's biggest scandalHow Square made its own iPad replacementFour reasons we don’t have flying cars—yetYou can now live out Westworld with your Amazon EchoHow Oprah’s network finally found its voiceLooking for more? Sign up for our daily newsletter and never miss our latest and greatest storiesRead More

Leave A Comment

More News


NDTV Auto - Latest

ABR - News

Super Street RSS Feed

Automotive News Breaking

All Motor Trend Stories RSS

Automotive news - from

Disclaimer and is not the owner of these news or any information published on this site.