DIY meets autonomous driving

When you get a team of professional engineers in a group doing projects for fun, you can get some pretty awesome making going on. Michael Worry at Nuvation Engineering clued me in on a pretty unique DIY project in the realm of autonomous driving.

After-hours at Nuvation, engineers take on exciting projects for fun at a program called Nuvation Garage.

"During the day we do engineering because we're paid for it and after-hours we do engineering for the love of the game," Worry says.

One Nuvation Garage project they dubbed "DiscoFish."

It's a 65-foot long, 38,000-pound "mobile dance party" vehicle in the shape of an anglerfish that the Nuvation team takes to the art festival and experimental community event Burning Man in the Black Rock Desert in northern Nevada. Other features include a pro audio sound system, 3,000 individually addressable light-up LED scales, a dance stage, a 20-foot flamethrower, and an 11 kW searchlight visible from four miles away.

"We have office parties on it, and we have a lot of fun with it," Worry says (I think this is an understatement).

What's more, it's an autonomous vehicle.

"One of the things we wanted to do very early on was to use it as a platform in order to advance autonomous research, because I really wanted to be able to have a beer and have an art car at Burning Man," Worry says. If you aren't familiar with Burning Man art cars, do a Google image search and you'll find some other near large-scale maker projects – it's better to see them than to read about them.

Worry leveraged Steve Waslander, Assistant Professor, Department of Mechanical and Mechatronics Engineering at the University of Waterloo, with whom Nuvation has partnered for research and development of collaborative autonomous vehicle systems, to develop DiscoFish's autonomous driving capabilities.

Figure 1: Nuvation's DiscoFish autonomous IP diagram. Image courtesy of Nuvation. All rights reserved.
(Click graphic to zoom)

It has an Ubuntu Linux computer running Robot Operating System (ROS), motors on the gas, brake, and steering wheel of the vehicle, LIDAR in front and a GPS unit on the roof. These systems are hooked up to two screens accessible to the driver; one displays the forward-looking LIDAR data being used for obstacle avoidance, and one shows the GPS-managed waypoint route that DiscoFish is following. "You can reach above your head and tap on the screen, 'I want to go here, here, and here' and click 'GO' and the vehicle starts turning and peels off and follows the route," Worry says. "It's super fun!"

The environment at Burning Man includes thousands of people moving about on foot and on bicycles, plus other art cars. It provides excellent examples of some of the more challenging and unpredictable conditions in which an autonomous vehicle may need to maneuver. Worry says it's taught them a lot about how people might use autonomous driving capabilities.

"I found there were times where I wanted to manually control the gas and brake while keeping the steering autonomous so DiscoFish would keep following the pre-programmed route," Worry says. "There were other times where I could see a mild obstacle that I wanted to steer around, but I wanted it to not fall out of its navigation, so I would briefly take it out of autonomous steering and nudge around an obstacle, then let it continue on its way. So we started to discover all these different use cases that were really interesting."

Limitations to the system also became apparent in this real-world environment, where humans and their autopilots would have to work together to solve problems not yet solvable by today's computer-vision and sensor-based systems.

"We have an obstacle avoidance program that looks at how far away the next obstacle is, and then can modify its speed all the way down to zero to make sure it never hits anything," Worry says. "There are so many people milling around that most of the time the computer says 'I don't have a clear route so I'm just going to sit here' where as a human I can say 'I'm going to just nudge forward six inches and I'm pretty sure these people will notice and get out of a way.' As humans we are very good at accurately predicting, and can even influence, the behavior of people around our vehicle while driving, but those sorts of circumstances are certainly tougher for a computer to take on because they can't make eye contact with a pedestrian and let him know to please step aside and so forth. It's been really interesting to see these sorts of use cases involve what it's going to look like to have increasing autonomous capabilities in our vehicles."

It's not your usual maker project, but it sounds very cool to play with and at the same time it's advancing research in the exciting area of autonomous driving. Keep on making with a purpose, Nuvation Garage!

See DiscoFish in action and read more technical details about the system's design.

Images courtesy of Nuvation Engineering. All rights reserved.