Human drivers average about 1 fatality per 100 million vehicle miles traveled. [1] So we are still a long ways from being able to access safety in comparison to humans.
[1] https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/...
I think you are begging the question - are cars more dangerous than horses? Just from reading random history, it seems like a lot of famous people were killed by riding accidents. And even today when the vast majority of people don't ride horses, I bet you can name a famous person who had a terrible accident.
Personally, I hope it never happens. The US is still suffering from when its cities were redesigned for cars, vs. public transit and pedestrians.
Self-driving cars can and will play a very important role in transportation, but cities should be designed around people, not cars.
Edit: I'm by no means criticizing your hope, and I do agree with you that cities should be redesigned to "accommodate" self driving cars, such as "buffer garages" (where autonomous cars are parked in a FIFO way, not in the way cars are parked today), auto-charging stations, smart curbs for pick-up/drop-off, etc.
Taking the SF Bay area an example: Sure I can take the Caltrain or the Bart to get to nearly any city but it'll always be the long way around.
Think San Mateo to Union City. Crossing a bridge it's direct. The road infra structure is already there and the cities are built around it. Sure there's probably some cryptic bus route but the difference right now is (with a quick google) 35 mins vs 1.5 hr.
I think self driving cars have the potential to build on our car centric ecosystems. Look at how Uber/Lyft have adapted into Uber Pool and Lyft Line. They've done more to change people's driving habits than any changes to the public transportation system. I'm excited to see how self driving cars solve this by dropping the price point further and exploring larger pools.
that is the use case best suited for flying tub https://www.youtube.com/watch?v=EQK9m_OBVgY
But the problem is that car-centric systems are not scalable and sustainable.
Sure, self-driving cars would do well in the Bay Area, but they would only make traffic worse, as now people are willing to spend a longer time in traffic, since they are no longer driving.
Self-driving cars actually remove one of the negative feedback mechanisms of traffic.
What's so special about efficient routing with a driver-less car? There's no route that a human driver couldn't follow just as well if assisted by GPS software.
Putting people on a train is going to be far more efficient.
For example, accidents that occur at 35 MPH or less are much less likely to result in a fatality or major injury, due to the amount of kinetic energy that a human body can safely dissipate. So if Google cars have even slightly better braking or speed control, you're going to see an improvement. Looking at the average speed at which accidents occur would be useful information.
When the industry or the press discuss the safety of self-driving cars, they almost always do it in terms of fatality rates. Not general accident rates.
The CA DMV data has enough info that you can distinguish between "driver had to take over to avoid crash" vs. "vehicle was unable to proceed due to problem ahead and needed help". The first number needs to get very low before self-driving can work. The second, Waymo proposes to handle by having a very limited remote control capability, so remote support can get a vehicle past a strange problem.
That's an element of right-wing policy created to fight safety regulation. The position is that since the severity of non-fatal workplace injuries is hard to measure, only fatal accidents should be recorded.[1][2] Then there's too little data to make policy.
[1] https://edworkforce.house.gov/uploadedfiles/testimony_michae... [2] https://www.oshalawblog.com/2017/04/articles/its-official-os...
Mainly because good driving is often synonymous with defensive driving. And you don't need to drive defensively if there's a large number of cars on the road which won't unexpectedly change lanes, turn right from the left lane or wildly change speed for no reason. If anything, human drivers might start doing unsafe maneuvers more as they see SDVs around them react quicker and safer to their bad driving. Why bother getting in the correct lane when you can pick the less congested one and rely on SDVs to move out of your way when you barge past?
There's also the issue of people switching from SDV to manually operated vehicles. What if someone uses an autonomous car for commuting on weekdays and gets used to the car acting for them, then uses a normal car on the weekends? They might expect the car to emergency brake for them when a pedestrian steps out into the road, and in the split second it takes them to realise the car will not do that for them they've already traveled too far to stop in time.
https://orfe.princeton.edu/~alaink/SmartDrivingCars/Papers/R...
the question is at what point we decide that it’s “good enough” to be a life saving compromise
humans are really terrible drivers compared to sensors. the number of inputs, precision, and quick decision making are all much better for driving than humans
Assuming simulation is proven to be good enough for training cars, then having the ability to simulate crashes so that cars can respond in an optimal way is going to make a massive difference.