I wonder if the most advanced autonomous vehicles already have those sensors installed and how well the signals are integrated into the prediction and decision making processes.

Quick calculation: Lets think about people who drive their whole lives, not city folks. Lets think about work hours driving as it's likely to be the most by quantity, and vacation or pleasure driving should be roughly equivalent.

30 minutes each average way 5 days a week starting at 16, so by 40 they've driven 5/7365 = 260hrs / year. 260 24 years = 6240 hours.

8 Hours a day, 5 days a week, with salaried level vacations is 2000 hours a year. Many drivers are likely to work way more than that. Anyway, 6240 / 2000 = 3.12 years. That means that if you started driving at 18, you'd have the same hours driving as a 40 year old.

This also means that by the gladwell definition of 10,000 hours, you don't become an expert driver till 54.5 years.

However a pro driver becomes and expert after 5 YEARS of driving.

This also means that only 22% (those over 55) of america is likely to be an expert driver. http://www.infoplease.com/us/census/data/demographic.html

If HN population is the same as americans (it's likely skewed older), then 78% of hacker news isn't a gladwellian "Expert".

So self driving cars don't need to be better than most people they need to be better than a group of people who are very likely to be experts.

1) Average driver drives ~13.5K miles/year (http://cars.lovetoknow.com/about-cars/how-many-miles-do-amer...) with an average speed of ~32 mph (http://www.ridetowork.org/transportation-fact-sheet). This translates into 420 hours/year. Or a person becoming an expert driver by the age of 40, if we accept the "10,000 hours rule"...

2) But the Gladwell's "10,000 hours rule" was proven wrong (http://journals.sagepub.com/doi/abs/10.1177/0956797614535810). For example, if we follow FAA that recently changed the rules to require 1500 hours for ATP (i.e. airline pilots) license, then you are talking about 3.5 years (for average driver) vs. 9 months (for taxi driver). Not that big of a difference.

This means that on average the US driver becomes an "expert" by the age of 21-22. This actually matches quite well with the insurance company rates that drop drastically at about this age.

The majority of car accidents are correlated with someone's fault, related to altered state, inexperience, distraction, or bad judgment. Some types of people are prone to accidents, others are less prone.

But in the self driving car world it's all equal - it's completely arbitrary who is a victim of an accident.

Someone who has never run a red light or had an accident in their life because they are an alert, careful, experienced, defensive driver suddenly has all that stripped from them and now they are driving at just an "above average" level.

Why should the good, consistent drivers be punished? (And they absolutely would if we settle for just "better than average"). I'm all for driverless cars, but only with patience; no one should even entertain mass adoption of self-driving cars without a minimum of multiple years of testing in live traffic with zero accidents or major traffic violations (by fault of self-driving car).

Seeing the traffic violations at the start of the SF program, we probably ought to expect the mass adoption minimum 5 years out, more likely 10-15.

And that, my friend, is a very good thing. We slowly raise the bar and eventually get rid of bad drivers - because in the future only people with a near perfect driving record will be granted a license.

I did say "much better than average" I agree that if "just above average" this would be a fail.