7 ms·

Because this sort of auto->manual transition of control does not work [1]. 1: https://www.amazon.com/Our-Robots-Ourselves-Robotics-Autonomy-ebook/dp/B00SI02AWK

Because this sort of auto->manual transition of control does not work [1]. 1: https://www.amazon.com/Our-Robots-Ourselves-Robotics-Autonomy-ebook/dp/B00SI02AWK https://www.amazon.com/Our-Robots-Ourselves-Robotics-Autonom...

studentrob 10y ago
Any notable quotes or stats from the book? I'd guess some of Tesla's stock value is tied up in the idea that the auto-manual transition of control does work.
- oconnore 10y ago
  I lent out my copy, but one particularly relevant section compares results from the aviation industry's use of pure-autopilot and HUD-augmented landings. Pilots maintain their flying skills, and get better overall results by staying in the loop with an augmented vision system. Good pilots often avoid full-autopilot entirely, or are forced to abort only seconds from touchdown, resulting in poorer landings. Bad pilots over rely on autopilot, and are therefore under-trained to deal with exceptional situations, resulting in incidents like the infamous Air France crash. https://youtu.be/4nDdqGUMdAY?t=183 https://youtu.be/4nDdqGUMdAY?t=183
  - studentrob 10y ago
    Cool. Tesla should review this book and talk. Paraphrasing a few quotes from your link ... "The most difficult challenging tech problem is not full autonomy, but rather the interaction between automated systems and humans" The subtitle of the book is also notable: "Robotics and the Myths of Autonomy" The speaker almost seems to be arguing that full autonomy is impossible. If we start up a system, then its behavior has still been kicked off and wrapped by humans. Interesting perspective. I'd like to read the book myself after listening to a bit of this talk. Thanks!
    - ghaff 10y ago
      It's not so much that full autonomy is impossible. We "fire and forget" many type of systems with no possibility of human intervention--at least not in a timely manner. What you can't do is start an automated process, comforting yourself that an inattentive human can always take over in a split-second if the automation messes up.
      - studentrob 10y ago
        > We "fire and forget" many type of systems with no possibility of human intervention--at least not in a timely manner Practically speaking, I agree with what you're saying. For autonomous systems as we think of them today, humans can set them and walk away. Philosophically, if I throw a ball into space, is its continued movement autonomous? Kicking off a computer program is just a complex ball. We don't have true randomness in computers. > What you can't do is start an automated process, comforting yourself that an inattentive human can always take over in a split-second if the automation messes up. I 100% agree with this. I think that's a fundamental flaw in Tesla's current plan to achieve autonomous vehicles. It's also a huge liability risk, for which they are not insured, to be dishing out 15,000 systems like this every month. At least the other consumer-available self-driving car systems require hands on the wheel. Tesla doesn't even seem to do that.
        Dylan16807 10y ago
        > We don't have true randomness in computers. Yes we do.
        studentrob 10y ago
        Computers are deterministic. Maybe you can convince yourself that something in the outside world isn't, and generate numbers based off of that, like radioactive decay. Even that may be deterministic. I think it's a question for quantum-mechanics or philosophers. Anyway, practically speaking, we tend to be happy with random numbers that other people would have a really difficult time copying.
  - studentrob 10y ago
    Interesting. I didn't realize humans have already attempted so many different automated systems. In the talk you linked, he points out how, historically, systems are never fully automated because humans are not comfortable with that. He seems to predict that cars will not be the first systems that we allow to become fully autonomous, and admits he could be wrong. It's surprising to me that he feels texting should be permissible while driving with a driver-assist system [1]. His whole argument here doesn't add up for me. I'll have to check out his book. It sounds like he supports driver-assist mechanisms, though maybe without Tesla's beta release schedule of OTA updates. He seems unconvinced that full autonomy will ever be achievable. I think I still believe in Google's plan, despite the shadow he casts on full autonomy. Zero deaths and full disclosure seems better than selective reports and the possibility for more fatal accidents. Given there are now a decent number of investors on both sides, I imagine the debate will continue as to which is the best path: testing driver-assist in consumer vehicles, or testing full autonomy with company-controlled cars. I'd like to hear what Mindell thinks would be the best hand-off from computer to driver in a driver-assist vehicle such as Tesla's. He answers a question about that generally here [2], but is mainly talking about airplanes where there is a chance for a smoother transition. How does one slow down time such that a hand off is possible in a vehicle during an adverse event that the car cannot handle itself? I feel that is the most pertinent question to today. It seems like the car needs to see the adverse event coming, but since it is not ready to handle the event, the car is unlikely able to give the driver much extra time. By definition of adverse event, it would now seem to have been better if the driver had been paying complete attention the whole time. [1] https://youtu.be/4nDdqGUMdAY?t=41m32s https://youtu.be/4nDdqGUMdAY?t=41m32s [2] https://youtu.be/4nDdqGUMdAY?t=35m37s https://youtu.be/4nDdqGUMdAY?t=35m37s
    - oconnore 10y ago
      When he talks of humans being richly involved in the environment, not sleeping in the trunk, I imagine an interface with several levels. Example: you are driving down a highway that has full monitoring by overhead cameras. No obstacles are going to jump out at you, and the position of the car is mapped precisely in real time. The human controls via a joystick to select a lane position, and brakes are automatically applied if avoidance is needed. Now the car leaves the highway, onto an isolated road. Before leaving the known environment, the interface switches. Now the lane position becomes a trajectory projection, and the driver must observe the road. The task of safely following the trajectory is automated. The interface adds random noise to the trajectory projection, to test that the driver is paying attention. Failure to demonstrate control results in the car pulling off to a rest stop. Finally the car goes downtown. A well known road, but with many pedestrians and bicycles. The interface is based around identifying obstacles, with a joystick for max safe speed <-> stop. Trajectory is automated, with the driver choosing turns. The car scans for pedestrians, but if the driver does not acknowledge them, the car slows since the driver is not paying attention. If the car misses a pedestrian, the driver is aware and has been conditioned and trained to react in time. All of these interfaces would require driver attention (minimal on the highway), but would significantly reduce fatigue associated with driving.
cpncrunch 10y ago
Actually, that's precisely how (real) autopilots work. If the autopilot can't reliably continue flying (e.g. in extreme turbulence, severe icing, or failure of some part of the system), it will immediately disengage with a loud warning sound, and the pilot has to take immediate control.
- kalleboo 10y ago
  Specifically because this is such a precarious transition of control, pilots have extensive training in staying alert and being prepared to take over, have co-pilots, and have rules about adequate rest times etc. And yet there are still air accidents involving improper use of the Auto Pilot. We can expect none of these things from the average driver - they're more likely to be watching Harry Potter on their iPad
  - ghaff 10y ago
    Furthermore, I imagine that an auto going down a road is far more likely to detect (or not detect) a situation where a human needs to take action right now. Aircraft have systems to detect approaching objects and things like sensor failures usually aren't going to require split-second action. (e.g. the Air France 447 crash involved multiple pilot errors over many minutes rather than a failure to take over rapid enough control.)
- greendestiny 10y ago
  If planes had to fly within centimetres of hard barriers there would be no autopilot.
  - kbenson 10y ago
    If planes had to fly within centimeters of hard barriers there would be no pilots.
  - Dylan16807 10y ago
    I trust a computer much more than a human to keep a tiny gap at high speed for hours on end. Also if you have a nice clear barrier, it becomes easier to automatically manage steering.

studentrob 10y ago

Any notable quotes or stats from the book? I'd guess some of Tesla's stock value is tied up in the idea that the auto-manual transition of control does work.

oconnore 10y ago
I lent out my copy, but one particularly relevant section compares results from the aviation industry's use of pure-autopilot and HUD-augmented landings. Pilots maintain their flying skills, and get better overall results by staying in the loop with an augmented vision system. Good pilots often avoid full-autopilot entirely, or are forced to abort only seconds from touchdown, resulting in poorer landings. Bad pilots over rely on autopilot, and are therefore under-trained to deal with exceptional situations, resulting in incidents like the infamous Air France crash. https://youtu.be/4nDdqGUMdAY?t=183 https://youtu.be/4nDdqGUMdAY?t=183
- studentrob 10y ago
  Cool. Tesla should review this book and talk. Paraphrasing a few quotes from your link ... "The most difficult challenging tech problem is not full autonomy, but rather the interaction between automated systems and humans" The subtitle of the book is also notable: "Robotics and the Myths of Autonomy" The speaker almost seems to be arguing that full autonomy is impossible. If we start up a system, then its behavior has still been kicked off and wrapped by humans. Interesting perspective. I'd like to read the book myself after listening to a bit of this talk. Thanks!
  - ghaff 10y ago
    It's not so much that full autonomy is impossible. We "fire and forget" many type of systems with no possibility of human intervention--at least not in a timely manner. What you can't do is start an automated process, comforting yourself that an inattentive human can always take over in a split-second if the automation messes up.
    - studentrob 10y ago
      > We "fire and forget" many type of systems with no possibility of human intervention--at least not in a timely manner Practically speaking, I agree with what you're saying. For autonomous systems as we think of them today, humans can set them and walk away. Philosophically, if I throw a ball into space, is its continued movement autonomous? Kicking off a computer program is just a complex ball. We don't have true randomness in computers. > What you can't do is start an automated process, comforting yourself that an inattentive human can always take over in a split-second if the automation messes up. I 100% agree with this. I think that's a fundamental flaw in Tesla's current plan to achieve autonomous vehicles. It's also a huge liability risk, for which they are not insured, to be dishing out 15,000 systems like this every month. At least the other consumer-available self-driving car systems require hands on the wheel. Tesla doesn't even seem to do that.
      - Dylan16807 10y ago
        > We don't have true randomness in computers. Yes we do.
        studentrob 10y ago
        Computers are deterministic. Maybe you can convince yourself that something in the outside world isn't, and generate numbers based off of that, like radioactive decay. Even that may be deterministic. I think it's a question for quantum-mechanics or philosophers. Anyway, practically speaking, we tend to be happy with random numbers that other people would have a really difficult time copying.
- studentrob 10y ago
  Interesting. I didn't realize humans have already attempted so many different automated systems. In the talk you linked, he points out how, historically, systems are never fully automated because humans are not comfortable with that. He seems to predict that cars will not be the first systems that we allow to become fully autonomous, and admits he could be wrong. It's surprising to me that he feels texting should be permissible while driving with a driver-assist system [1]. His whole argument here doesn't add up for me. I'll have to check out his book. It sounds like he supports driver-assist mechanisms, though maybe without Tesla's beta release schedule of OTA updates. He seems unconvinced that full autonomy will ever be achievable. I think I still believe in Google's plan, despite the shadow he casts on full autonomy. Zero deaths and full disclosure seems better than selective reports and the possibility for more fatal accidents. Given there are now a decent number of investors on both sides, I imagine the debate will continue as to which is the best path: testing driver-assist in consumer vehicles, or testing full autonomy with company-controlled cars. I'd like to hear what Mindell thinks would be the best hand-off from computer to driver in a driver-assist vehicle such as Tesla's. He answers a question about that generally here [2], but is mainly talking about airplanes where there is a chance for a smoother transition. How does one slow down time such that a hand off is possible in a vehicle during an adverse event that the car cannot handle itself? I feel that is the most pertinent question to today. It seems like the car needs to see the adverse event coming, but since it is not ready to handle the event, the car is unlikely able to give the driver much extra time. By definition of adverse event, it would now seem to have been better if the driver had been paying complete attention the whole time. [1] https://youtu.be/4nDdqGUMdAY?t=41m32s https://youtu.be/4nDdqGUMdAY?t=41m32s [2] https://youtu.be/4nDdqGUMdAY?t=35m37s https://youtu.be/4nDdqGUMdAY?t=35m37s
  - oconnore 10y ago
    When he talks of humans being richly involved in the environment, not sleeping in the trunk, I imagine an interface with several levels. Example: you are driving down a highway that has full monitoring by overhead cameras. No obstacles are going to jump out at you, and the position of the car is mapped precisely in real time. The human controls via a joystick to select a lane position, and brakes are automatically applied if avoidance is needed. Now the car leaves the highway, onto an isolated road. Before leaving the known environment, the interface switches. Now the lane position becomes a trajectory projection, and the driver must observe the road. The task of safely following the trajectory is automated. The interface adds random noise to the trajectory projection, to test that the driver is paying attention. Failure to demonstrate control results in the car pulling off to a rest stop. Finally the car goes downtown. A well known road, but with many pedestrians and bicycles. The interface is based around identifying obstacles, with a joystick for max safe speed <-> stop. Trajectory is automated, with the driver choosing turns. The car scans for pedestrians, but if the driver does not acknowledge them, the car slows since the driver is not paying attention. If the car misses a pedestrian, the driver is aware and has been conditioned and trained to react in time. All of these interfaces would require driver attention (minimal on the highway), but would significantly reduce fatigue associated with driving.

cpncrunch 10y ago

Actually, that's precisely how (real) autopilots work. If the autopilot can't reliably continue flying (e.g. in extreme turbulence, severe icing, or failure of some part of the system), it will immediately disengage with a loud warning sound, and the pilot has to take immediate control.

kalleboo 10y ago
Specifically because this is such a precarious transition of control, pilots have extensive training in staying alert and being prepared to take over, have co-pilots, and have rules about adequate rest times etc. And yet there are still air accidents involving improper use of the Auto Pilot. We can expect none of these things from the average driver - they're more likely to be watching Harry Potter on their iPad
- ghaff 10y ago
  Furthermore, I imagine that an auto going down a road is far more likely to detect (or not detect) a situation where a human needs to take action right now. Aircraft have systems to detect approaching objects and things like sensor failures usually aren't going to require split-second action. (e.g. the Air France 447 crash involved multiple pilot errors over many minutes rather than a failure to take over rapid enough control.)
greendestiny 10y ago
If planes had to fly within centimetres of hard barriers there would be no autopilot.
- kbenson 10y ago
  If planes had to fly within centimeters of hard barriers there would be no pilots.
- Dylan16807 10y ago
  I trust a computer much more than a human to keep a tiny gap at high speed for hours on end. Also if you have a nice clear barrier, it becomes easier to automatically manage steering.