The video above shows a very specific use case of a common object (cars, animals) getting in the way of a vehicle and the vehicle braking, or trying to swerve around the object. It's true that's better than a distracted driver talking on their cell phone, but it doesn't prove that self driving cars are better than human drivers from a general purpose perspective.

The video that I posted is avoiding an uncommon round about with the sun present in the background. This is a scenario that we would expect all humans to be able to perform without any difficulty at all, even if they have never driven on that road before. However, the self-driving vehicle fails to successfully navigate the round abouts because today's artificial intelligence is narrowly scoped. Humans still greatly outperform computers when performing this type of analysis. Analysis that is very common and easy for us to do. For instance, we expect a baby at the age of 18 months to climb a set of stairs, regardless of the depth or material of the stairs, however, for a robot to do this, it was a major feat, just 5 years ago.

http://robotics.umd.edu/news/news_story.php?id=6655

I love the idea of technical advancements in AI, but I think people are WAY to bullish on the timeframes involved to make those advancements possible.