|Developer Diary · You Heard It Here First · Sunday 14 March 2004|
|DARPA Grand Challenge Reveals Machine Limitations|
As I predicted back in December the DARPA Grand Challenge ended in complete defeat for the robotic participants. Ditches (as predicted) and anti-ditches ("berms") proved to be the most formidable problem. A machine cannot tell the difference between a ditch and a berm and a shadow even in non-moving laboratory conditions; forget about real-time streaming video recognition. The favorite car in the race, the Carnegie Mellon entry, got hung up on a berm and continued spinning its wheels until they caught fire--illustrating another problem with robots: their inability to cope with unexpected situations.
A human would think, "Hmm, I'm not moving forward. I seem to be stuck. I'll trying backing up at different angles." For a machine this kind of thinking is difficult. The Red Team car apparently did not even know it was stuck, much less solve the problem. It did not even realize there was a problem. Maybe if the team put temperature sensors in the wheel wells they could have got a signal: "Warning! Tires about to ignite."
A lot of machines got disqualified in the pre-trials for colliding with objects. The Red Team car got past this by crawling at 5 mph as it slavishly followed GPS waypoints on a microscopic scale. I guess it didn't have that berm in its 100-gijillabit topography map of the desert. GPS can only take you so far. Fundamentally it requires real intelligence to size up the situation on the ground and negotiate it.
I saw a dog agility contest in Quebec City in February. Imagine border collies leaping over hurdles, slaloms, ladders and rings as fast as their legs can carry them. Now that's obstacle avoidance!
Frequently radar is cited as one of the voodoo sensors that has some kind of special power to detect nearby objects. The technology used is radar ranging which determines the distance along a path of an emitted signal. The same thing can be done with sound (sonar) or light (lidar). The problem is that in the rough-and-tumble circumstances of an off-road race the vehicle is bouncing up and down and the image from the ranger will be chaotic. To use radar effectively requires a stable platform.
So it's back to the drawing board for the DARPite robotic vehicles. I wonder how far they will get next year?
|return to John Chamberlain's home · diary index|
|Developer Diary · about · email@example.com · bio · Revised 14 March 2004 · Pure Content|