Mercedes 2013 100 mile journey through all sorts of traffic without touching the wheel and including roundabouts, traffic lights, autobahn and crowded city centres with loads of pedestrians and so on is still the most impressive demo I have ever seen:
https://www.youtube.com/watch?v=it-3HmDFWa0

They are now going to do it with much more production ready equipment.

Tesla collecting who knows what data, and processing it who knows how, no matter how many miles it covers with unknown comprehensiveness, does not impress me.

The only thing that they have shown is a readiness to release beta software in safety critical applications to wholly untrained members of the general public, in cars capable of 'ludicrous' acceleration.

They are the only car company which has killed people AFAIK with their AP.

As usual Davemart is spreading false and negative news about Tesla on the internet just like a paid fossil fuel lobbyist would do.

There is only one confirmed death accident in a Tesla with autopilot activated, not many, as claimed by Davemart. And he also lies about other automakers having no death accidents in cars with their autopilot activated. We don’t know how many deaths other automakers’ autopilot systems may have been involved in as their cars are not collecting that info yet. Tesla is the only car maker in the world that systematically record autopilot info and upload it to Tesla’s data center for further analysis for every car they sell. Tesla is also the only car in the world that has a black box that record this info and that can be used to recover this info in case the car is destroyed so much in the accident that its connection to Tesla’s data center is lost and the info cannot automatically be uploaded to Tesla.

Then there is the question of how many lives Tesla’s autopilot is saving. People are being killed in traffic mainly because of human error from being inattentive with or without autopilot. However, with autopilot on the chance that a driver is inattentive to the point that it causes a death is far less because the autopilot is always attentive and can drive the vehicle safely in most situations. We need billions of miles with autopilot on before conclusive evidence can be made on the safety advantages of Tesla’s autopilot. However, that data will come by 2018 when over 500,000 Tesla cars are driving on autopilot most if not all of the time in all kind of situations.

My view is that you will kill a few, (possibly quite a few) people along way to fully autonomous vehicles. You have to balance this with the number of lives you will save when you get it more or less working. The problem is that we will know the identity of the people killed by AVs, but will only see the lives saved statistically - no-one will know whose lives were saved, so no credit will be given for this.
The trick will be to pace the development such that minimum people are killed or injured and yet it happens as quickly as possible.
You may have to treat each new city as a new environment, certainly at the start as new things are encountered (cyclists and bike lanes in SF, for instance).
After a while, we may plateau and see no substantially new phenomena when we encounter a new city. (And eventually, new countries).
What you do not want is to move to too many new cities all at once so that the new situations come in faster than they can be programmed (or trained up).

Mahonj is right. You cannot avoid accidents and deaths with fully autonomous vehicles. Neither during their development nor after they are developed for deployment on a massive scale. Accidents will still happen no matter how much we try to avoid it just like they do with human drivers. Both hardware and software can fail and other human drivers can crash into self driving vehicles. The objective is not to avoid accidents but to minimize the chance that they will happen. The data from large fleets of autonomous cars will show how successful the automakers are at minimizing the number of accidents per mile traveled in various circumstances. We need hundreds of thousands of fully autonomous cars to make scientific sense from that data in a practically short time like on a monthly or quarterly basis. Tesla will not start to get that before the end of 2018. Other automakers will start to get it by 2021 to 2022.

Very good points mahonj. Accepting a few more death can be a small price to pay for the many more lives that are saved from getting faster to market with this life saving tech.

I don’t think that 1000 will be lost on that account. Tesla could make the level 5 software as soon as dec 2017 (although I expect several months of delays) and I doubt we will see more than 10 new deaths until then in Tesla cars with autopilot on. There could even be zero new death. However, Tesla is not alone and others may see some death before they get a fully working version of the software ready. Still 1000 is a lot IMO. It is probably less than 100 before every car maker has developed their own level 5 systems.

In principle poor countries could get self-driving cars just as fast as rich countries because self-driving taxies will cost less per miles than self owned low cost gassers cost per mile. However, rich countries will be willing to pay a large premium to be first in line for self-driving cars as they are safer and save time. Initially the demand will far outstrip the capacity to make the self-driving cars. Tesla will probably see much higher share prices when they succeed being the first auto company that has fully self driving cars. Self driving ability is going to be the best selling point ever. It trumps everything else by a large margin.