Victorville (CA) – At first glance the blue Volkswagen Passat looks like any other car, well if you excuse the spinning laser sensors, numerous corporate logos and a big white number “3” plastered on the driver side windshield.  But the windshield obstruction isn’t a big deal because this car from Stanford University doesn’t need a driver and from what we’ve seen at the DARPA Urban Challenge qualifications in Victorville this week, the robotic car “Junior” performs just as well – if not better – than most human drivers we’ve seen.  Stanford is a sure favorite to complete and possibly win Saturday’s race and despite being the top dog, team members were more than happy to share their hardware and software secrets.

Stanford tells reporters the technology behind "Junior" 

Team leader Sebastian Thrun explains the advantages of using lasers versus cameras 

Team leader Sebastian Thrun says breaking the California vehicle code is allowed in extreme cases 

Junior's early testing against an inflatable Santa 

Unlike the previous Grand Challenge races where robots raced in the desert, the Urban Challenge requires vehicles to navigate around a mock city environment and this means curbs… lots of curbs.  Software lead Michael Montemerlo said detecting curbs isn’t as simple as aiming a laser to figure out height changes.  The main sensor on Junior is a spinning Velodyne laser that gives a 360 degree point cloud map of the surrounding environment.  Montemerlo told reporters that the car does sense height changes, but detects curbs by following rings of points as they get closer and farther.  “This works much better than simply firing off a laser to sense just height,” he said.

Race participants will also have to detect moving cars which presents more difficulties.  Since the Velodyne sees objects as a “blob” of points it is hard to differentiate between a car and a mailbox, according to Montemerlo.  The secret here to detect occlusion by figuring out the empty space as a car moves.  Stationary objects, like mailboxes and concrete walls, should have no occlusion as Junior drives by.  Junior further detects cars by superimposing a roadmap as a filter to the laser data.  “The roadmap shows where cars shouldn’t be,” Montemerlo said.

During the race, dynamic obstacles will be placed quickly in front of the vehicles and Stanford had a unique way of testing Junior’s avoidance programming against such “quick movers”.  Montemerlo told fellow team members to find something “soft” so that an impact wouldn’t hurt the car and the students found a six-foot tall inflatable Santa on a motorcycle from a local hardware store.  “It looks a little strange to have a big Santa in a parking lot in the middle of summer,” Montemerlo joked.

The plan was to pull the Santa in the path of the robotic car with the help of a rope.  The first try wasn’t successfully, as you can see in the video, and Junior hit the hapless Santa.  After some reprogramming, Junior detected the Santa and broke sharply to a stop.  After a few seconds, the car recognized the Santa a stationary roadway obstacle and drove around it.

Team Stanford has now qualified for the Urban Challenge race and will no longer have to do any qualification runs.

Comments (3)

Show/Hide comment form