I don't think these cars have a human mind to have road rage.
About the CBS 60 Minutes segment, I no longer watch much TV so missed it. The full video requires subscription to view, but I found a free page with a transcript of it. It is at:
Hands off the Wheel - CBS News.
The transcript describes exactly why I think Google has a more advanced or at least more interesting technology. It is because they put AI (artificial intelligence) into their prototype.
Bill Whitaker: There are so many variables, so many different scenarios. How is it possible to put all of that knowledge into a car?
Chris Urmson: And that's really the trick, right? And that's what makes this hard. You can't just kind of go through and enumerate, you know, the thousand different scenarios it might encounter, 'cause it's not 1,000. There's an infinite number of them, right? And so the trick is to develop these algorithms that can generalize.
By generalize, he means "think" and this is how it works. The algorithms are trained to recognize other cars, pedestrians, cyclists, and animals from their movements, size, and shape. Each car's daily driving experience is analyzed, uploaded and shared. The cars can then make predictions and choices based on the collective knowledge of the fleet. Look in the lower left corner as one of Urmson's cars encounters a pickup truck that stops to parallel park.
Bill Whitaker: Now, how does the computer know that it's someone intending to back into a parking space, and not someone who's just stopped in the street?
Chris Urmson: Our cars have seen thousands and thousands of vehicles. And they get a feeling, you know, they get a feeling really for what the behavior of those vehicles are going to be.
Bill Whitaker: Really?
Chris Urmson: So its seen lots of cars backing up and so it understands if there's a space here, and a car stopped just in front of it, that means it's gonna probably back into that spot.
AI inference software takes huge amount of computing power. I believe that's the reason Google car is limited to 25 mph to give the computer time to "think", and also for the test driver to take over.
What are other limitations that these researchers are working on now? This comes from that 60 Minutes transcript.
Right now, the technology can't handle snow. Google's cars can't operate in heavy rain. The Mercedes S500 can't decipher hand gestures from traffic cops or pedestrians. Four million miles of roads in the U.S. must be mapped in ultra high-definition detail. The automakers call these solvable problems. In the meantime, the car industry plans to automate the driving experience feature-by-feature, what some are calling revolution-by-evolution.
Sometimes a sensor can have a basic limitation that's insurmountable because of the laws of physics. I recall an experiment many years ago with LIDAR. A research was conducted to apply it for wire strike avoidance by low-flying aircraft. It was found that the laser would scatter off old dull wires and let them be "seen". New and shiny wires would reflect the laser in a direction that may not hit the receptor back at the aircraft, hence would not be "seen". I do not know if they could ever overcome that.
The public is usually enthusiastic about new technological advances because it is not told of difficulties, caveats, or limitations. Experts know about the problems they face, but will always say that they can solve it. Maybe they will. Maybe they cannot. It's just not a sure thing.
