The Google car is in Washington, DC, and the Antiplanner managed to hitch a ride around downtown. My host, Anthony Lavandowski–sometimes driving, sometimes just sitting in the driver’s seat–answered a number of questions about the car.
The Cato Institute’s David Boaz stands next to the Google Prius. In addition to the spinning laser sensor on top of the car, note the infrared sensors in the front bumper (there’s a similar one in the center of the back bumper). The laser sensor finds nearby objects while the infrared sensors can detect objects much further away. Click on any photo for a larger view.
He said the car and hardware cost about $100,000, but Google has just a handful of them. When they go into mass production, he estimated an ordinary car could be retrofitted for a couple of thousand dollars. Some cars already have many of the sensors the Google car uses, so the cost of retrofitting such cars would be much lower.
Inside the car are more sensors, including a camera next to the rear-view mirror that detects traffic signals. The “driver” of the car is a desktop computer in the trunk; the laptop only allows the operators to monitor what the car sees.
As we threaded through downtown DC traffic, he noted that the car relies on GPS for only the most general purposes. Mainly it relies on the information it senses and its built-in knowledge of the area. For example, Google programs the location and height of every traffic signal the car might encounter so it knows where to look for the signals.
The small white boxes are pedestrians; the larger boxes are other motor vehicles. Anthony is pointing to a traffic light. Other lines show traffic lanes and crosswalks. The upper righthand corner gives the current status, including the legal speed limit and traffic signal.
People don’t like the idea of robots killing people, Anthony wryly noted as the car lurched to a halt when a pedestrian crossed in front of us, so Google programmed the car to drive extra cautiously. He promised they would eventually reprogram the brakes so the stops wouldn’t be so abrupt. Someone honked when several other vehicles took advantage of the car’s good nature to merge in front of us; eventually, they’ll probably program the car to be a bit more aggressive in such situations.
I asked him what they did about the fact that most drivers drive 5 to 10 mph over speed limits. The car has two modes, he said. If you’re going 75 mph and you put it in one mode, it will maintain that speed where it is safe to do so, slowing down as appropriate for curves, traffic, and other appropriate reasons. If you put it in the other mode, it will strictly obey all traffic laws, getting into the righthand lane if everyone else is going faster. No doubt they will eventually have “plus-5-mph” modes or whatever to take into account that many roads are signed for slower than the speeds people actually drive.
Anthony told me a couple of surprising things. No state actually outlaws driverless cars, he said; traffic codes were written before anyone conceived of self-driving cars, so the laws only require that a licensed driver be in the driver’s seat, not that they actually be operating the car. Google’s goal is to get the states to affirmatively recognize that driverless cars exist and eventually may operate without a licensed driver. Nevada and Florida have already passed such laws and Google hopes California will soon follow.
Operators of modern cars aren’t really driving them, he noted. When the operator turns the wheel or put a foot on the brake, the car gets a signal and then responds to that signal using built-in software. In some situations, many cars will override the signals they get from the operators for safety reasons. Self-driving cars just introduce one more layer of software between the operator and the car.
I asked him whether liability laws would be a barrier. Automakers probably won’t bring self-driving cars to market under current laws, he speculated–but those laws won’t stop Google. He promised that Google would stand by its software: If a self-driving car is involved in an accident, the car will have a record of what other vehicles involved were doing up to the accident. If the other vehicle is at fault, the record will prove it, and if Google’s car is at fault, Google would pay the cost and fix the problem so it won’t happen again.
So the main thing left is improving the systems so the they are ready for prime time. The car I rode in has detailed information about Washington DC built in–but it can’t drive itself back to California without a lot of new data entry and programming. Perhaps one of the reasons why Google photographs all the streets in the country is to help build the database for its driverless cars. Despite the need for more work, seeing the car in action makes it seem much closer to reality.