To gain control of the car he has to do one of three things: hit a red button near his right hand, touch the brake or turn the steering wheel. He did so twice, once when a bicyclist ran a red light and again when a car in front stopped and began to back into a parking space. But the car seemed likely to have prevented an accident itself.
I'm not sure I'm ready to trust my life to "seemed likely".
Now, I'm not the "average" driver. I've had several courses in advanced driving techniques, ranging from the Air Forces High Speed Driving course at Lackland AFB to the "Offensive Driving" course (protective services driving) course at Quantico. I'm also an experienced amateur race driver (sports car and rally). I have three years experience in Law Enforcement, and 6 years as a third shift cab driver, and have seen my share of "statistics" (what we call idiot drivers who kill themselves or others). Even so, I'm not of the opinion that this technology will be ready in a mere decade. While a robot can drive better than some drivers most of the time, there are too many situations where trained human driver will surpass a robot. Here in the US, too many drivers simply aren't trained well - it's far too easy to get a license.
I have a couple problems with this. I live in northern Wisconsin. While a computer may have sensors that allow it to determine it's current traction, it can't anticipate the traction it will have a couple hundred yards down the road. Not a problem in southern California, but essential around here four months out of the year. Nor will that computer be able to anticipate a deer's movement the way a human can - computers don't have intuition, nor can any amount of programming give it intuition. A snow storm will blind a computer's visual sensors - even infra red, so unless roads are wired bad weather will stop a robot car. Even rain will render electronic sensors useless - "shine" will make road lines invisible (assuming the lines are maintained, which is far from certain) and will eliminate any temperature differential between pavement and shoulder.
Because the robot cars would eventually be less likely to crash, they could be built lighter, reducing fuel consumption.
Which means that any accident you do have will be far more likely to kill you.
Sure, the human could take over in case of a problem, but people will simply trust the robot and pay no attention to the situation. They won't even notice there is a problem until metal gets bent.
As for robotic cabs, there is far more to being a cab driver than getting from point A to point B. A robotic cab won't help an old lady load her groceries into the car, or help her get them into her house. It won't know which bar is busy tonight so you can have a good time, or know which restaurant is likely to have a long wait for a table. It won't be able to figure out where the incoherent drunk lives.
True Story #1: I once picked up a Japanese man at the Madison airport who was going to Ft Atkinson (small town about 40 miles away) for buisiness. He was supposed to have been met at the airport, but wasn't. He didn't know what hotel he was supposed to be staying at. I drove him to Ft Atkinson, and after entering town turned off the main road. I don't know why I chose that corner to turn at. A few blocks later I turned again, for no apparent reason. I saw a hotel - and passed it up. I saw a second hotel and pulled in. It was the correct hotel - he was a day earlier than expected, which is why no one was there to meet him. Let's see a robot do that.
True Story #2: I picked up a man who had a bandage on his arm and was bleeding. He flagged me down - he had no phone to call an ambulance. I'm a trained EMT and recognized that it was arterial blood (from the color). I applied a tourniquet and drove him to the nearest hospital (which was 15 minutes closer than the one he originally asked for). The ER doctor told me my actions saved his life. A robot won't be able to do that either.
I like technology as much as the next guy, but I'm not blind to it's limitations. A computer can "mimic" a humans reactions only in circumstances it has been programmed to deal with - and no programmer will think of everything. It will be a long time before artificial intelligence will be able to make good decisions based on very incomplete data nearly as well as the human brain can. Computers don't have intuition - they can only fake it in circumstances that they are specifically designed to handle. Computers are far better than humans when it comes to repetitive tasks, or situations it has been programmed to deal with. They are no match for the human brain when it comes to dealing with new problems and a mixture of incomplete and erroneous data.
Most of the tech used here would be great - FLIR would certainly come in handy for seeing deer or pedestrians in poor visibility. Automatic brakes which activate when sensors detect an obstacle wouldn't hurt - far too many people tailgate. In ideal conditions, a robocar may be a safer driver than me, but in adverse conditions I am certain that I am safer driving myself - and ideal conditions are few and far between.
I also question how much a self driving car is going to cost. How long before the politicians (who can afford these vehicles) decide that humans should no longer be allowed to drive themselves - leaving lower income people unable to get to work and further impoverishing the poor?