[osrec]

I disagree with your initial sentiment. In my opinion, we can have self driving cars without a large human toll. I just think we need to stop trying to merge self driving cars into a road system designed for human operators. Moreover, we should not be "beta testing" our self driving cars on roads with human operators. Accidents will happen, as ML models can and do go unstable from time to time. Instead, we should look to update our roads and infrastructure to be better suited to automated cars. Till then, I hope those martyred in the name of self driving technology are not near and dear to you (even if you'd feel it's worth it).

[tedivm]

To me beta testing should be a long period of time where the computer is running while humans are driving, with deviations between what the computer would do if it had control versus what the human actually does being recorded for future training. The value add is that the computer can still be used to alert to dangerous conditions, or potentially even overriding in certain circumstances (applying break when lidar sees an obstacle at night when the human driver didn't see it).

The problem is that Uber needs self driving cars in order to make money, and Tesla firmly believes that their system is safer than human drivers by themselves (even if a few people who wouldn't have otherwise died do, others who might have died won't and they believe those numbers make it worth it).

[Splines]

It's surprising that this isn't the standard right now. I'm certain the people at Tesla/Uber/Waymo have considered this - I'm curious why this approach isn't more common.

[kolinko]

It was happening. Many companies, Tesla including, gathered a lot of data for training the system.

The problem is that you cna only learn so much without actual practice.

[turtlebits]

I'm sure it is the standard during testing. I doubt you can do this for the general population.

[kbutler]

How about all those martyred by prolonging our current manual driving system for the years band decades it will take to roll out separate infrastructure for vehicles no one owns because they can't drive them anywhere?

I think we need to keep the human driver in control, but have the computers learning through that constant, immediate feedback.

And get rid of misleading marketing and fatal user experience design errors.

[buvanshak]

>but have the computers learning through that constant, immediate feedback.

I don't know what is stopping them from simulating everything inside a computer.

Record the input from all the sensors when a car equipped sensors is driven through real roads by a human driver. Replay the sensor input, with enough random variations and let the algorithms train on it.

Continue adding to the library of sensor data by making the sensor car by driving it through more and more real life roads and in real life situations. Keep feeding the ever increasing library of sensor data to the algorithm, safely inside a computer.

[kolpa]

>I don't know what is stopping them from simulating everything inside a computer.

Obviously they've already tried that, and it doesn't work. The map is not the territory.

In theory, there is no difference between theory and practice. In practice, there is.

[buvanshak]

Not following you here. What do you mean by "The map is not the territory."..

What I mean is that. Do not "teach" the thing in real time. Instead collect the sensor data from the cars a human is driving (and also collect the human input also), and train the thing on it, safely inside the lab.

You say, they have done it already. But I am asking if they have done it enough. And if that is so, how come the accidents such as these are possible, when the situation is pretty out of a text book in basic driving?

[osrec]

I don't think it'll take years to update our infrastructure. For example, we could embed beacons into catseyes to make it easier to know where the road boundaries are etc. Also, we could make sections of the highway available with the new infrastructure piece by piece. It is just as progressive as your suggestion, but the problem becomes a whole lot easier to solve when you target change towards infrastructure as well as the car itself.

[kbutler]

These will take a lane from regular traffic and give it to rich people who buy an expensive car.

It also will not ensure exclusion of manual vehicles, so it won't create the exclusion necessary for the predictable driving environment.

[pbhjpbhj]

In think it could work, that lane can be the same as is used for autonomous lorries (that potentially benefit everyone if they deliver cheaper goods).

[osrec]

Not really, the road would still be fully usable by a standard car.

[kbutler]

"I just think we need to stop trying to merge self driving cars into a road system designed for human operators. Moreover, we should not be "beta testing" our self driving cars on roads with human operators."

Sounds like requiring exclusive access - I apologize if that was a misinterpretation.

If you have human and automated drivers in the same roads, the computers have to be able to cope with the vagaries of human drivers.

How can you then get away from '"beta testing" our self driving cars on roads with human operators' if that is their deployment environment?