5 ms·

Also millions of miles of fleet learning a day, which Uber doesn't have, since they don't own the cars. Self-driving cars is a supervised learning problem, and

Also millions of miles of fleet learning a day, which Uber doesn't have, since they don't own the cars. Self-driving cars is a supervised learning problem, and he who has the data wins

argonaut 10y ago
It is a supervised learning problem, but that doesn't mean it will be solved by just shoving large amounts of data into a black box algorithm. This is not MobileEye's approach, and it almost certainly is not Tesla's approach. The data is useless unless it is annotated (e.g. a human labels where the lanes are, where the obstacles are, what the bicylist is doing, etc.) - that's the bottleneck, not collecting large amounts of raw sensor data + driver actions.
- tansey 10y ago
  > The data is useless unless it is annotated (e.g. a human labels where the lanes are, where the obstacles are, what the bicylist is doing, etc.) - that's the bottleneck, not collecting large amounts of raw sensor data + driver actions. Except that is exactly what Nvidia did, and it worked out fine for them: https://arxiv.org/abs/1604.07316 https://arxiv.org/abs/1604.07316
  - argonaut 10y ago
    Nope. That car just does lane keeping - it doesn't even do turns or lane changes. This is all stuff solved 20 years ago. And even then it only achieves 98% autonomy on lane keeping - this is a task that needs 100% accuracy. You should not be running into the median every couple miles. Furthermore, they augmented the data with left/right-offset cameras to supplement the data with examples of "bad" camera views. This is not present on Tesla cars (because these sensors are only used for training purposes) In fact, the paper actually supports my point. They collected all this data for one task, lane keeping. They subdivided the problem of autonomous driving, and managed to solve one small subproblem (the easiest subproblem of autonomous driving, solved for decades already). They avoided the need for annotators, but only because they used specialized purpose-built cameras to augment the data.
    - bennyg 10y ago
      > They collected all this data for one task, lane keeping. They subdivided the problem of autonomous driving, and managed to solve one small subproblem (the easiest subproblem of autonomous driving, solved for decades already). They avoided the need for annotators, but only because they used specialized purpose-built cameras to augment the data. So why not several autonomous subsystems that use specialized purpose-built cameras and don't need annotators? I'm not saying that like it's easy - obviously it's not. Just seems scalable.
      - argonaut 10y ago
        The solution was specific to that subproblem. The left/right-offset cameras were for the sole purpose of providing examples of what it would look like if the car was deviating off path. The same trick would not work for any other problems. Can you think of similar camera data augmentation tricks for obstacle detection, drivable path segmentation, bicyclist signaling/intention, pedestrian detection, and so on?
lallysingh 10y ago
There's also the matter of that laser scanner on some cars - usually costing more than the car...
- dpc59 10y ago
  http://www.bloomberg.com/features/2015-george-hotz-self-driving-car/ http://www.bloomberg.com/features/2015-george-hotz-self-driv... people are trying to fix this problem too
- nl 10y ago
  Not anymore. Complete LIDAR units are already available for under $500[1]. There are other, cheaper sensors which aren't as good as full LIDAR but are available for under $100[2]. There are plenty of other options hitting the market soon too[3] [1] http://www.teraranger.com/products/teraranger-lidar/ http://www.teraranger.com/products/teraranger-lidar/ [2] https://www.pulsedlight3d.com/ https://www.pulsedlight3d.com/ [3] https://www.washingtonpost.com/news/innovations/wp/2015/12/04/the-75000-problem-for-self-driving-cars-is-going-away/ https://www.washingtonpost.com/news/innovations/wp/2015/12/0...
  - Animats 10y ago
    It's not quite here yet. I've been following this since the DARPA Grand Challenge, when we had to use a huge SICK LMS just to get a line scanner. There are now several affordable indoor line scanners. Outdoor sunlight-tolerant systems cost more. 3D scanners, which scan in multiple planes, are still expensive. There are some MEMS devices coming along. Flash LIDAR will probably win out in the end, once someone does the sensor IC development to get the price down from $100K. Back in 2003, I dragged a VC down to see Advanced Scientific Concepts in Santa Barbara. They make the best flash LIDAR. But they were happy being a DoD and aerospace contractor, selling expensive one-offs. The Dragon spacecraft uses an ASC flash LIDAR to dock with the space station. DARPA buys their units. But their price point is around $100K. There's no inherent reason it has to be that expensive, but it takes custom sensor ICs made in small quantities. Last March, Continental AG (German tire/brake/auto parts company) bought the technology from ASC.[1] We'll have to see how that works out. This is the right technology if the price point can be brought down. [1] http://www.spar3d.com/news/lidar/flash-lidar-company-acquired-by-german-auto-parts-supplier/ http://www.spar3d.com/news/lidar/flash-lidar-company-acquire...
intrasight 10y ago
Data isn't that hard to gather. Uber has more robotics scientists than all the others combined.
- agumonkey 10y ago
  Funny when it started as a GPS tracking webapp.
- oh_sigh 10y ago
  Even google(with boston dynamics?)

Also millions of miles of fleet learning a day, which Uber doesn't have, since they don't own the cars. Self-driving cars is a supervised learning problem, and he who has the data wins

argonaut 10y ago

It is a supervised learning problem, but that doesn't mean it will be solved by just shoving large amounts of data into a black box algorithm. This is not MobileEye's approach, and it almost certainly is not Tesla's approach. The data is useless unless it is annotated (e.g. a human labels where the lanes are, where the obstacles are, what the bicylist is doing, etc.) - that's the bottleneck, not collecting large amounts of raw sensor data + driver actions.

tansey 10y ago
> The data is useless unless it is annotated (e.g. a human labels where the lanes are, where the obstacles are, what the bicylist is doing, etc.) - that's the bottleneck, not collecting large amounts of raw sensor data + driver actions. Except that is exactly what Nvidia did, and it worked out fine for them: https://arxiv.org/abs/1604.07316 https://arxiv.org/abs/1604.07316
- argonaut 10y ago
  Nope. That car just does lane keeping - it doesn't even do turns or lane changes. This is all stuff solved 20 years ago. And even then it only achieves 98% autonomy on lane keeping - this is a task that needs 100% accuracy. You should not be running into the median every couple miles. Furthermore, they augmented the data with left/right-offset cameras to supplement the data with examples of "bad" camera views. This is not present on Tesla cars (because these sensors are only used for training purposes) In fact, the paper actually supports my point. They collected all this data for one task, lane keeping. They subdivided the problem of autonomous driving, and managed to solve one small subproblem (the easiest subproblem of autonomous driving, solved for decades already). They avoided the need for annotators, but only because they used specialized purpose-built cameras to augment the data.
  - bennyg 10y ago
    > They collected all this data for one task, lane keeping. They subdivided the problem of autonomous driving, and managed to solve one small subproblem (the easiest subproblem of autonomous driving, solved for decades already). They avoided the need for annotators, but only because they used specialized purpose-built cameras to augment the data. So why not several autonomous subsystems that use specialized purpose-built cameras and don't need annotators? I'm not saying that like it's easy - obviously it's not. Just seems scalable.
    - argonaut 10y ago
      The solution was specific to that subproblem. The left/right-offset cameras were for the sole purpose of providing examples of what it would look like if the car was deviating off path. The same trick would not work for any other problems. Can you think of similar camera data augmentation tricks for obstacle detection, drivable path segmentation, bicyclist signaling/intention, pedestrian detection, and so on?

lallysingh 10y ago

There's also the matter of that laser scanner on some cars - usually costing more than the car...

dpc59 10y ago
http://www.bloomberg.com/features/2015-george-hotz-self-driving-car/ http://www.bloomberg.com/features/2015-george-hotz-self-driv... people are trying to fix this problem too
nl 10y ago
Not anymore. Complete LIDAR units are already available for under $500[1]. There are other, cheaper sensors which aren't as good as full LIDAR but are available for under $100[2]. There are plenty of other options hitting the market soon too[3] [1] http://www.teraranger.com/products/teraranger-lidar/ http://www.teraranger.com/products/teraranger-lidar/ [2] https://www.pulsedlight3d.com/ https://www.pulsedlight3d.com/ [3] https://www.washingtonpost.com/news/innovations/wp/2015/12/04/the-75000-problem-for-self-driving-cars-is-going-away/ https://www.washingtonpost.com/news/innovations/wp/2015/12/0...
- Animats 10y ago
  It's not quite here yet. I've been following this since the DARPA Grand Challenge, when we had to use a huge SICK LMS just to get a line scanner. There are now several affordable indoor line scanners. Outdoor sunlight-tolerant systems cost more. 3D scanners, which scan in multiple planes, are still expensive. There are some MEMS devices coming along. Flash LIDAR will probably win out in the end, once someone does the sensor IC development to get the price down from $100K. Back in 2003, I dragged a VC down to see Advanced Scientific Concepts in Santa Barbara. They make the best flash LIDAR. But they were happy being a DoD and aerospace contractor, selling expensive one-offs. The Dragon spacecraft uses an ASC flash LIDAR to dock with the space station. DARPA buys their units. But their price point is around $100K. There's no inherent reason it has to be that expensive, but it takes custom sensor ICs made in small quantities. Last March, Continental AG (German tire/brake/auto parts company) bought the technology from ASC.[1] We'll have to see how that works out. This is the right technology if the price point can be brought down. [1] http://www.spar3d.com/news/lidar/flash-lidar-company-acquired-by-german-auto-parts-supplier/ http://www.spar3d.com/news/lidar/flash-lidar-company-acquire...

intrasight 10y ago

Data isn't that hard to gather. Uber has more robotics scientists than all the others combined.

agumonkey 10y ago
Funny when it started as a GPS tracking webapp.
oh_sigh 10y ago
Even google(with boston dynamics?)