On Campus: Potholes, Pedestrians and Parking Lots

Cars that drive themselves are certainly cool, but cars that could avoid potholes would be a godsend in a city like Pittsburgh.

With the latest Collaborative Research Lab (CRL) between General Motors and Carnegie Mellon, researchers aim to refine the autonomous driving technologies that were so spectacularly put to use in Boss, winner of the 2007 DARPA Urban Challenge. That competition between driverless vehicles took place over 55 miles of urban and suburban roads.

"The Urban Challenge was fantastic, but you couldn't use Boss to get to work in the morning," says Chris Urmson, a research scientist in the Robotics Institute. "This lab will deal more generally with pedestrians, traffic--the wide variety of situations you encounter in a normal day. It will produce a much more broadly capable vehicle."

The five-year, $5 million CRL will support a range of projects in the emerging field of driverless vehicles, such as advanced motion planning and next-generation software and hardware architecture.

Research in the CRL will carry forward many of the concepts from the Urban Challenge, and aims to improve upon them. For example: Potholes.

"Boss didn't really understand potholes at all," says Urmson. "Had it understood potholes, it would understand that it could straddle them."

Also, Boss had a necessarily simple model of what obstacles and vehicles were around it, and could sometimes get confused about whether they were moving or not. "We want to look at how we can better represent other vehicles," Urmson says.

The autonomous driving CRL is directed by Raj Rajkumar, professor of electrical and computer engineering. In addition to Urmson, Robotics Institute faculty involved in the lab include Senior Systems Scientist John Dolan and Systems Scientist Paul Rybski.

Other projects to be investigated by the team include cars that can detect bicycles and better navigate parking lots.

When it comes time to park, the car will use aerial images (such as those available in Google Maps) to know where the parking spots are. "Most of the time when you or I are in parking lot, it's much like driving on the street," Urmson says. "You know which lane to stay in — We'd like to use imagery to provide the vehicle with the structure of the parking lot, and therefore make the vehicle drive more like a person." Indeed, having a car that drives like a person is the ultimate goal of this CRL.

Imagine a car that could navigate by landmarks--like "turn left after you pass the Carnegie Museums." That was one of the ideas that emerged from the first GM-Carnegie Mellon CRL, which focused on information technology and human-vehicle interaction.

In that collaboration, and associated spinoff research projects, SCS and Human-Computer Interaction Institute (HCII) researchers also worked on cars that could recognize signs and read them in real time--even translate them into different languages.

Faculty involved in human-vehicle interaction projects include Daniel Siewiorek, director of the HCII and Buhl University Professor of SCS and ECE; Jie Yang, senior systems scientist in HCII; and Jodi Forlizzi, the A. Nico Habermann Chair in SCS and associate professor in the School of Design and HCII.

Another project, which used hand gestures to control vehicle navigation and music, can be seen in a video available in the "Master's Students" section of the HCII Web site, www.hcii.cmu.edu.

Judging from past collaborations between GM and CMU, this CRL will produce new technologies that will radically change how we drive. "It's going to be exciting," says Urmson. --Karen Hoffmann (S'04) is a Pittsburgh-based freelance writer who covers science, health, and the environment.
For More Information: 

Jason Togyer | 412-268-8721 | jt3y@cs.cmu.edu