Some very interesting news from Google today:
Google shows off winning image-recognition system, likely to assist in autonomous car efforts
http://9to5google.com/2014/09/08/goo...s-car-efforts/
To summarize the article, Google's computers are getting very good at interpreting what images are.
A quick recap: There are two prevailing ideas about how Autonomous Cars should work: Google's, and everybody else's. Google's method (thus far) has been heavily reliant on pre-mapped routes (mapping every detail, such as how tall the curbs are and where the sewer covers are to within inches) and Lidar, which is an ultra-expensive radar-esque system that uses lazers instead of radio waves. All of this is meant to generate a large 3D model in which the computer can navigate.
The other approach is to use the many cheap cameras and sensors already on most new cars, only have a super-smart computer analyse the images, thus forgoing the expensive Lidar sensors and detailed maps.
That's what makes Google's advancement with interpretive software so interesting. Could it be that Google is realizing that Lidar is perhaps a little too much - that an easier solution exist?
Who knows. Google, obviously isn't telling.
I am a fan of the optics method. Before I get to why, though, here is another news story:
Verizon CEO: Autonomous Vehicle Could Happen In 3-5 Years
http://detroit.cbslocal.com/2014/09/...-in-3-5-years/
Now, you may rightly ask what Verizon ought to know about Autonomous Cars. I did. The answer is that Verizon is interested in the communications between vehicles and infrastructure (V2I) that, they suppose, must preclude autonomous cars. If you read the article, you will find Verizon talking about chips embedded in roadway striping and cameras mounted on bridges and buildings adjacent to Highways in order to create 'Smart Highways.' And of course they would want that, because they are a communications and network company. If Autonomous cars are developed along that route, they would stand to benefit greatly.
My problem with this is, autonomous cars that require chips in the road or bridge-mounted cameras to coordinate their lane changes aren't really autonomous, are they? In the same way that Google's car now needs extensive 3D maps in order to drive, these cars are limited to environments built for them, only in this case physically instead of digitally. The cars are not the ones making the decisions, but the infrastructure is. That is bad.
It's easy to think that cars will develop that way, reliant on infrastructure, because so far almost every mode of transportation we have involves dumb vehicles and smart infrastructure. Trains follow switches and signals and cars are required to follow stoplights and painted lines on the asphalt, even when these things are not efficient because the networks get bogged down.
One of the revolutions that will be brought on by autonomous cars is the switch to smart vehicles and dumb infrastructure - or at least minimally involved infrastructure. Autonomous cars (and in the future, even autonomous buses trains) will know their destinations, the traffic conditions around them, and all the routes available to them. The car will make the choice, not the infrastructure. The cars will negotiate among themselves how best to use all the available pavement. There will be no need for painted lines or coordinated stop lights. Pedestrian crossings will still exist, sure, but they will work differently - perhaps not requiring cars to stop, but merely drive around pedestrians crossing the street. The cars will be fully autonomous in that they will communicate with each other and not need the supervision of infrastructure to govern what they do.
Anything short of that will be a handicap.
Just my thoughts.