Another Tesla on autopilot steers towards a barrier • Hacker News Board

Re: Another Tesla on autopilot steers towards a barrier

#21

Post by makomk » Fri, Apr 06, 2018, 1:19 PM UTC

darkerside wrote: post #11

I wonder if Tesla has a way to visualize how disruptive a change to Autopilot. They've supposedly got all this driving data lying around. Would they be able to apply the updated algorithms to that data and quantify how many decisions would be affected, and by how much? i.e. 30% of past decisions would have been changed with the new algorithm, 5% changed by an angle of over 2 degrees. On a similar note, shouldn't cust…

I don't think they have that much data that's useful for this purpose. They may have billions of miles of some kinds of data, but they only upload occasional samples of the camera video that would be required to test a feature of this sort. I can't find the exact link now, but someone reverse-engineered it and it was something on the order of a few seconds a few times an hour.

Re: Another Tesla on autopilot steers towards a barrier

#22

Post by mv4 » Fri, Apr 06, 2018, 1:19 PM UTC

Musk wanted to regulate AI to combat existential threat; in reality, it's not super intelligent self-aware AI, but buggy/shitty AI that's trying to kill people.

Re: Another Tesla on autopilot steers towards a barrier

#23

Post by maxerickson » Fri, Apr 06, 2018, 1:20 PM UTC

fairpx wrote: post #13

Autopilot tech in general should be open and shared among ccompanies and universities. The more brains we can put on this the better imho. If these things happen too often, Im afraid regulators will start making laws that prevent the use of this tech, which I believe has the potential to really make life better for all... one day.

I think (at least in the US), the relationship the automakers have with regulators is such that if the automakers think they can make money by selling a feature, they will be able to get the regulators to come up with a way to allow it.

Not necessarily one as simple as OTA updates whenever the manufacturer feels like it.

As far as regulations preventing use of the tech, if level 2/3 systems end up causing more crashes than human drivers, we shouldn't allow use of them.

Re: Another Tesla on autopilot steers towards a barrier

#24

Post by donkeyd » Fri, Apr 06, 2018, 1:21 PM UTC

GoToRO wrote: post #8

ProblemFactory wrote: post #4

If Tesla autopilot relies on good quality road markings, then it's not yet usable in 80% of the world.

It's not usable anywhere. What if I draw a fake line on the road? will it follow it? what if you have wet paint and another vehicle smudges it diagonally to the lane?

I have a pretty simple lane keeping aid in my car that keeps me within the lines. Yesterday it suddenly steered towards a wall. I think it may have been confused by either glare (low sun right in front of me) or a shadow line on the road.

Re: Another Tesla on autopilot steers towards a barrier

#25

Post by maxerickson » Fri, Apr 06, 2018, 1:22 PM UTC

donkeyd wrote: post #17

darkerside wrote: post #11

I wonder if Tesla has a way to visualize how disruptive a change to Autopilot. They've supposedly got all this driving data lying around. Would they be able to apply the updated algorithms to that data and quantify how many decisions would be affected, and by how much? i.e. 30% of past decisions would have been changed with the new algorithm, 5% changed by an angle of over 2 degrees. On a similar note, shouldn't cust…

How do you define a 'decision' in something like autopilot though? Also, if I remember correctly, only a couple of images per second are sent to Tesla. If there's a fraction of a second where it may decide to run into something, it may not actually appear in test results.

One way is to extract the drive path for some given sensor input. Then you can classify the drive path as safe, marginal, collision, severe collision, etc.

Changes from safe to safe aren't interesting. Changes in the wrong direction are obvious issues.

Re: Another Tesla on autopilot steers towards a barrier

#26

Post by fma » Fri, Apr 06, 2018, 1:22 PM UTC

fairpx wrote: post #13

Autopilot tech in general should be open and shared among ccompanies and universities. The more brains we can put on this the better imho. If these things happen too often, Im afraid regulators will start making laws that prevent the use of this tech, which I believe has the potential to really make life better for all... one day.

I have always wondered why each company is developing their own software and algorithms... How will NTSB certify the software? Or will they? I mean, they certify car safety features now through crash tests, roll over tests, etc... Why is Tesla, or anyone manufacturer, allowed to modify vehicle safety features through OTA updates without recertification?

I realize it adds more regulation...and I'm sure there can be a middle ground. But I foresee a scenario where an OTA update is pushed and the next morning there are car crashes everywhere.

Re: Another Tesla on autopilot steers towards a barrier

#27

Post by pipio21 » Fri, Apr 06, 2018, 1:22 PM UTC

One of the great things about self driving cars is that we are going to start "debugging Roads".

From my point of view all accidents until this time could be attributed to the civil engineer that build the road as much as the driver of the car.

You make a pedestrian pass between roads made only for aesthetic purposes so you put a sign forbidding the pass and call it done. But people of course use it and get killed.

You put concrete barriers between roads but make it so people could crash frontally against it, something you won't find in any European high speed road without deflectors that will make vehicles not crash against concrete.

Self driving will give us scientific evidence of what creates accidents like black boxes did with airplanes. Thanks to that we know that what looks like insignificant details like the color and placement of buttons, turns out to be essential.

Re: Another Tesla on autopilot steers towards a barrier

#28

Post by robotnixon » Fri, Apr 06, 2018, 1:23 PM UTC

taneq wrote: post #10

What does this say about the quality of the automatic emergency braking, if it can't detect a substantial metal object directly in front of the car?

Autopilot has trouble with stationary objects.

https://www.wired.com/story/tesla-autopilot-why-crash-radar/

Re: Another Tesla on autopilot steers towards a barrier

#29

Post by walrus01 » Fri, Apr 06, 2018, 1:23 PM UTC

mv4 wrote: post #22

Musk wanted to regulate AI to combat existential threat; in reality, it's not super intelligent self-aware AI, but buggy/shitty AI that's trying to kill people.

We will be killed by AI developers and the dunning kreuger effect, not armies of Skynet T-800s.

Re: Another Tesla on autopilot steers towards a barrier

#30

Post by DisruptiveDave » Fri, Apr 06, 2018, 1:23 PM UTC

fairpx wrote: post #13

Autopilot tech in general should be open and shared among ccompanies and universities. The more brains we can put on this the better imho. If these things happen too often, Im afraid regulators will start making laws that prevent the use of this tech, which I believe has the potential to really make life better for all... one day.

Serious question: how will it make life better for all? I haven't given that question a lot of thought, and I haven't heard legit answers from others yet.