Autonomous cars - what are the issues?

According to Google in the past seven years the company autonomous cars were involved in 11 minor accidents, for each of which the main factor was a human error. These data raise a number of issues. When autonomous cars will begin to enter the mass on the roads? What problems must be solved to make this happen? Fully autonomous there will be cars or human factor will continue to be essential?
More Info
$ 1 billion. For the creation of AI invest Elon Musk and other innovators
AI wait to replace the screens, and technology - to improve our senses
Video of the Day: This is what it looks like buying a car in 2020
Unmanned mini-buses run on the roads of the Netherlands
Better software
Only in the United States every 40 minutes one person dies as a result of a fatal crash. You might think that this is a lot, but actually, considering more than 200 million licensed driver appears that the fatal crash occurred about three million hours driving. Given the global statistics, the US is among the leaders in safe driving. Well, autonomous cars have to deal much, much better in this respect. And given existing software, it would be terribly laborious task. The reason for this is obvious - software laptops, smartphones, navigation and generally all modern devices can not operate for long periods without driving without crash, without rebooting. Similar problems are fatal in software autonomous cars.
Currently Google's self-governing cars avoid such problems by a person who remotely monitor the system and is ready at any time to react. But the system of protected critical issues and ideal software will as fundamentally needs to change the way software is written. In this line of thought, a curious fact is that when Boeing set a budget for the development of new aircraft, almost half of the money spent on the development and validation of software. And people have aircraft pilots.
Better navigation maps
Google's autonomous cars operate comfortably in the streets of Mountain View, California. But this is because the company has created Street View steroids real virtual world card. In this method, autonomous cars know not just any street, but just how they look when they are empty / full of people, cars and other obstacles. Without hiperdetaylni charts, self-governing cars will not be able to operate with its current sensor and calculating. For reference, similarly mapped under 10,000 kilometers. the road network of the United States. The total road network of the United States consists of 6.5 million. Km. roads.
Better sensors

Before people throw driver's licenses, autonomous car must be able to distinguish between dangerous and safe situations. Otherwise, the car will brake gives almost every meter of the route. For example, holes, broken glass and nails, are very difficult to spot, while actually not be crossed. While absolutely safe flying papers, bags and leaves, may seem dangerous suspicious car.
And this is just the beginning. The car must be able to decide whether pedestrian next to the walkway will cross or simply ignored. You must be able to assess the trajectory of other vehicles, including unpredictable cyclists and motorcyclists. And to make it even more complicated - autonomous cars must assess the sudden appearance of pedestrians, and respond to direct and indirect obstacles.

The human brain does a brilliant job in responding and sorting of events happening at the moment, but the spectrum of sensors on cars is still very poor. However, not even to mention what colossal calculations must be able to perform the software of the car. This was a hell of AI.
Communication system between vehicles.
To enter mass autonomous cars, they need to have a unified system of communication between themselves. Imagine the Internet, but not for human use rather than the cars themselves. With such a system, there is no need of self-governing car is smarter than man, since the emergence of different situations, the cars will be able to react to others around him.

On such a system is running, but its development is yet to come.

And last but definitely not least, come all ethical problems. Sometimes the driver must decide whether to turn left and kill yourself, turn right and kill three other people or beat brakes sharply, risking getting a motorcycle. This type of decision again boils down to AI. Moreover, this AI will need laws similar to those posed by Isaac Asimov (Three Laws of Robotics) in order to set to any situation and under-priority situation, and consequently to take and decision.

In conclusion

Autonomous cars with human operators will begin to appear on the streets in the next few years. But they will not become mainstream in the next 10 years, even less in countries such as Bulgaria. Fully autonomous cars will be able to take your child from home, to be brought to school and back home, something that can be visible only after mid-century.
However, this sector is extremely attractive and will continue to monitor developments with great interest.