4 ms·

That's what you think being more tuned in. Most of the people I talk to seem to think it's right around the corner. I think that's because Elon Musk does a go

by nscalf 7y ago

That's what you think being more tuned in. Most of the people I talk to seem to think it's right around the corner. I think that's because Elon Musk does a good job of making everyone hear him say it's nearly there. If anything, I think it's the opposite. I think they both see production ready AI and being almost here, and they don't have time to figure it out themselves.

Retric 7y ago
Full self driving may be a long way away, but driver assist like lane departure and automatic breaking have become major selling points. Further, in customers minds the features are binary rather than something where a slightly better version becomes a selling point. This makes joint ventures a win/win as they reduce the risk of being left behind while significantly reducing R&D costs.
simion314 7y ago
Agree, it can't be ready while you have to ignore static objects because the hardware and software is to weak to handle them IMO (please Tesla/Elon fans don't downvote an opinion without letting me know why you thing I am wrong and start a discussions, most of the downvotes I get are on my comments about Tesla's autopilot and Apple's hardware )
- Retric 7y ago
  From a car companies standpoint, good enough for many people to want to buy and use them is threshold, not good enough that you feel comfortable. Being what I would consider a beta tester for self driving cars seems a silly risk. But, people also pay lots of money driving vintage cars without modern safety systems which seems insane to me.
  - simion314 7y ago
    It can backfire, you need that instead of some random person killed by "autopilot" a bus with children, a family with children or some popular person gets killed, then you will get a ton of bad press like Boeing 747 MAX gets, everyone finding all the shorcuts you made and safety things you ignored for money.
    - toomuchtodo 7y ago
      When regulators have a problem with it, they can say so. At this time, regulators still allow Tesla to sell vehicles with Autopilot.
      - simion314 7y ago
        They allow it because of the fact you are supposed to pay attention 100% of the time, I am not saying this should not be allowed(having a system to force the drivers to pay attention would solve a lot of crashes in all cars) , my opinion is that ignoring all static objects means the hardware+software is not fast enough so there is a need for faster/better hardware to be able to correctly classify every object in front of a car
- donkeyd 7y ago
  So I've recently started driving a Model 3 and am a Tesla and Musk fanboy. That being said, I can still acknowledge things that are wrong. The major issue is people depending too much on the system. I think it's great that the 'autopilot nag' is much harsher than in the US (except for low speeds, in stop and go, it's flawless and nags too much here). The static object thing is an issue, yes, but it's massively overblown by people who tend not to drive a Tesla, because the driver can easily spot them (if they're paying attention) and apply corrective measures. I say this, because as of right now, Autopilot is a driving aid, not a replacement. (Yes, people have something to say about marketing, but that's a different discussion.) To me, the biggest issues with Autopilot (as it is supposed to be used today), are phantom braking and unexpected steering inputs. These require instant correction, are completely unpredictable and have the ability to scare other drivers or even cause crashes. Also, they happen a lot. Finding static objects on a highway is much less common than just driving on a straight road while being tailgated and therefore the impact is much bigger. Another thing is Autopilot unexpectedly disabling and the driver not being aware, because the indicator is not that clear. Sometimes it's a pleasant sound and if you don't look at the screen because you pay attention to the road, you might miss the visual indicator. Outside of the US, the major issue is that it's being built for the US and isn't fully adapted to non-US roads and rules. The 'Emergency Lane Departure Avoidance' feature (enabled by default before every drive) caused massive issues in the Netherlands on roads that have shared bike lanes. Also, in the Netherlands we have 'rush hour lanes', which are actually the hard shoulder, which is temporarily designated as a lane, as indicated by a green arrow over the lane. Autopilot doesn't know this and wants to move over, because it thinks the lane ends. To get back to your point, static object recognition is a small thing to overcome and is, so it seems to me, massively inflated by media that are heavily opinionated against Tesla. There's a much bigger problem in getting it working for every country and every weird situation. And there are things that humans do that I expect will be hard to pick up in an automated system. For example: spotting the driver next to you playing on the phone and passing for safety, seeing headlights reflected in a window and knowing a car is about to round a corner, etc.
  - simion314 7y ago
    > static object recognition is a small thing to overcome and is, so it seems to me, massively inflated by media that are heavily opinionated against Tesla Why do you think so? There will be often enough stopped cars on the road to cause issues, or cement barriers, construction sites. Do you think that having a filter that removes the objects with speed== 0 from the AI is something an safety engineer should do ? For me it looks like optimizations game developers do when your video card/CPU is not powerful enough so small objects are ignored at a larger distance. Driver issues are a different topic, since they signed that TOS they will get all the blame for killing themselves. IMO true self driving(without a human driver) will need more powerful hardware so no safety compromising shortcuts/optimizations are done(not even 1)

by nscalf 7y ago

Retric 7y ago

Full self driving may be a long way away, but driver assist like lane departure and automatic breaking have become major selling points. Further, in customers minds the features are binary rather than something where a slightly better version becomes a selling point. This makes joint ventures a win/win as they reduce the risk of being left behind while significantly reducing R&D costs.

simion314 7y ago

Agree, it can't be ready while you have to ignore static objects because the hardware and software is to weak to handle them IMO (please Tesla/Elon fans don't downvote an opinion without letting me know why you thing I am wrong and start a discussions, most of the downvotes I get are on my comments about Tesla's autopilot and Apple's hardware )

Retric 7y ago
From a car companies standpoint, good enough for many people to want to buy and use them is threshold, not good enough that you feel comfortable. Being what I would consider a beta tester for self driving cars seems a silly risk. But, people also pay lots of money driving vintage cars without modern safety systems which seems insane to me.
- simion314 7y ago
  It can backfire, you need that instead of some random person killed by "autopilot" a bus with children, a family with children or some popular person gets killed, then you will get a ton of bad press like Boeing 747 MAX gets, everyone finding all the shorcuts you made and safety things you ignored for money.
  - toomuchtodo 7y ago
    When regulators have a problem with it, they can say so. At this time, regulators still allow Tesla to sell vehicles with Autopilot.
    - simion314 7y ago
      They allow it because of the fact you are supposed to pay attention 100% of the time, I am not saying this should not be allowed(having a system to force the drivers to pay attention would solve a lot of crashes in all cars) , my opinion is that ignoring all static objects means the hardware+software is not fast enough so there is a need for faster/better hardware to be able to correctly classify every object in front of a car
donkeyd 7y ago
So I've recently started driving a Model 3 and am a Tesla and Musk fanboy. That being said, I can still acknowledge things that are wrong. The major issue is people depending too much on the system. I think it's great that the 'autopilot nag' is much harsher than in the US (except for low speeds, in stop and go, it's flawless and nags too much here). The static object thing is an issue, yes, but it's massively overblown by people who tend not to drive a Tesla, because the driver can easily spot them (if they're paying attention) and apply corrective measures. I say this, because as of right now, Autopilot is a driving aid, not a replacement. (Yes, people have something to say about marketing, but that's a different discussion.) To me, the biggest issues with Autopilot (as it is supposed to be used today), are phantom braking and unexpected steering inputs. These require instant correction, are completely unpredictable and have the ability to scare other drivers or even cause crashes. Also, they happen a lot. Finding static objects on a highway is much less common than just driving on a straight road while being tailgated and therefore the impact is much bigger. Another thing is Autopilot unexpectedly disabling and the driver not being aware, because the indicator is not that clear. Sometimes it's a pleasant sound and if you don't look at the screen because you pay attention to the road, you might miss the visual indicator. Outside of the US, the major issue is that it's being built for the US and isn't fully adapted to non-US roads and rules. The 'Emergency Lane Departure Avoidance' feature (enabled by default before every drive) caused massive issues in the Netherlands on roads that have shared bike lanes. Also, in the Netherlands we have 'rush hour lanes', which are actually the hard shoulder, which is temporarily designated as a lane, as indicated by a green arrow over the lane. Autopilot doesn't know this and wants to move over, because it thinks the lane ends. To get back to your point, static object recognition is a small thing to overcome and is, so it seems to me, massively inflated by media that are heavily opinionated against Tesla. There's a much bigger problem in getting it working for every country and every weird situation. And there are things that humans do that I expect will be hard to pick up in an automated system. For example: spotting the driver next to you playing on the phone and passing for safety, seeing headlights reflected in a window and knowing a car is about to round a corner, etc.
- simion314 7y ago
  > static object recognition is a small thing to overcome and is, so it seems to me, massively inflated by media that are heavily opinionated against Tesla Why do you think so? There will be often enough stopped cars on the road to cause issues, or cement barriers, construction sites. Do you think that having a filter that removes the objects with speed== 0 from the AI is something an safety engineer should do ? For me it looks like optimizations game developers do when your video card/CPU is not powerful enough so small objects are ignored at a larger distance. Driver issues are a different topic, since they signed that TOS they will get all the blame for killing themselves. IMO true self driving(without a human driver) will need more powerful hardware so no safety compromising shortcuts/optimizations are done(not even 1)