Why MIT’s ‘Duckietown’ uses adorable rubber toys to research self-driving cars

The “fictional startup’’ known as Duckietown at MIT aims to help researchers develop autonomous vehicle technology at a small scale.
The “fictional startup’’ known as Duckietown at MIT aims to help researchers develop autonomous vehicle technology at a small scale. –Courtesy MIT / CSAIL

In addition to making bath time lots of fun for millions of kids, rubber duckies may also help make self-driving cars a reality.

A small experiment at MIT is aiming to make autonomous car research less expensive, more accessible, and a lot cuter.

Duckietown is a small-scale model city housed at the university that was developed in a computer science class. Students were charged with developing a “duckiebot’’ capable of navigating the toy city’s traffic lanes on its own.

Each “duckiebot’’ costs about $150 to make. It includes a $35 microcomputer, a camera and a rubber duck. The ‘bots are programmed to identify lane dividers, road signs and traffic lights that are positioned along the small urban landscape.

Advertisement

Liam Paull, a postdoctoral researcher in MIT’s Computer Science and Artificial Intelligence Laboratory, says the program has transformed into a “fictional startup company.’’ Paull holds the equally fictional title of chief operating officer of “Duckietown Engineering.’’

Why rubber ducks? That’ an easy one, says Paull. They’re cute.

“We like ducks,’’ he said. “It’s a universally positive symbol. I don’t know anyone who has seen a rubber duck and not seen it as cute and fun.’’

When it comes to planning for self-driving cars, Paull says it helps to think small.

“We realized if you scale down autonomous driving to something very small there’s lots of research to do on a smaller scale with none of the logistical challenges of real autonomous vehicle research,’’ said Paull in an interview with Boston.com.

“I think there’s lots of research we can do in Duckietown that’s hard to do on a full-scale autonomous car,’’ he said.

One problem that has surfaced is the fact that sometimes the cameras will mistake a duck’s beak for a traffic cone. Visual mix-ups like could also happen in the real world. Paull says researchers are working on a solution to help the camera make the distinction.

Advertisement

Paull says the class is only offered to graduate students at the moment, but he and his team hope it will be expanded to undergraduate and high school students as well. The idea is to build an open source community that will help inform the development of self-driving cars.

“The dream would be that people adopt it and contribute back to it,’’ said Paull. “The more people get interested the better it will end up being.’’

Loading Comments...