Seeing the World Through Robot Eyes

Those of us who here at the DARPA Urban Challenge who haven’t built a robot car lately got a nice little seminar on the subject. The teacher: Stanford Racing Team software lead Mike Montemerlo. You can see a view of the way the team’s Volkswagen Passat TDI, called Junior, sees the world through its laser […]

Montemerlo_2
Those of us who here at the DARPA Urban Challenge who haven't built a robot car lately got a nice little seminar on the subject. The teacher: Stanford Racing Team software lead Mike Montemerlo.

You can see a view of the way the team's Volkswagen Passat TDI, called Junior, sees the world through its laser eyes over Montemerlo's shoulder.

The team likes lasers rather than camera-vision because they generate their own light sources rather than relying on fickle sunlight, with its constantly shifting shadows. They also work just as well at night as during the day.

To build up a continually-updated picture of the world, Junior's LIDAR units use the scattered light from their lasers to build up what Mike calls a 3D data cloud of rings at varying distances. As the rings shift positions with the cars movements, Junior can separate moving objects from stationary ones.

Junior_estop

To drive, Junior relies on rules about the relative cost of various maneuvers. How costly is a lane-change? Depends on if there's another vehicle in that other lane, where it is, and how fast it's moving.

For the Urban Challenge, team engineers will have Junior drive cautiously, though in after-hours simulations, they enjoy torquing it into the aggressive driving style they call Rambo Mode.

Stanford is considered by many the team to beat, since it won the Grand Challenge in 2005.

But victory is by no means a given. Even in E-Stop testing, the first of the trials the robot cars have been put through here at the qualifiers leading up to the Urban Challenge race itself, the Junior displayed the same kind of glitches experienced by other teams.

In this case, Junior's computers inexplicably stopped collecting Controller Area Network, or CAN, data that shows things like wheel position and other essential features of the car's state. Without the data, the car refused to budge after planned pause.

The minutes ticked buy while the engineers worked the problem. Finally, with only two minutes to spare before their time for testing was up, the rebooted the computer charged with collection the data, and Junior came back to life.