Brad Neuberg

@bradneuberg

Machine Learning Engineer. Research Scientist at SETI & NASA FDL. Previously @ Dropbox and Google. Started coworking. More:

San Francisco, CA
Vrijeme pridruživanja: svibanj 2008.

Tweetovi

Blokirali ste korisnika/cu @bradneuberg

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @bradneuberg

  1. prije 15 sati
    Prikaži ovu nit
    Poništi
  2. prije 15 sati

    “The flyby anomaly is a discrepancy between current scientific models and the actual increase in speed observed during a planetary flyby by a spacecraft. Spacecraft have greater speed than scientists had predicted, but thus far no convincing explanation has been found.”

    Prikaži ovu nit
    Poništi
  3. prije 20 sati

    CNNs need flat space and work best on 2D and 3D images. A subfield called geometric deep learning attempts to provide convolution operators for non-Euclidean spaces, such as the surface of a sphere: “An Idea From Physics Helps AI See in Higher Dimensions”

    Poništi
  4. prije 22 sata

    Really exciting work. Congrats to the Covariant team!

    Prikaži ovu nit
    Poništi
  5. prije 22 sata

    Nice photo of the team:

    Prikaži ovu nit
    Poništi
  6. prije 22 sata

    It also sounds like they’ve done true end to end learning of the entire robotics problem, against a single deep net. I wonder how they managed to have such a large network in the RL loop; traditionally deep RL nets are quite simple

    Prikaži ovu nit
    Poništi
  7. prije 22 sata

    They’ve created a single deep net that can actually work with different grippers! “We can condition a single neural network on both what it sees and the end-effector it has available. This makes it possible to hot-swap grippers if you want to.”

    Prikaži ovu nit
    Poništi
  8. prije 22 sata

    It sounds like there is a bit of imitation learning in the mix too: “But you don’t want to just use RL, because RL is notorious for taking a long time, so we bootstrap it off some imitation and then from there, RL can complete everything.”

    Prikaži ovu nit
    Poništi
  9. prije 22 sata

    I’d love to know more details about this: “Our system has few-shot adaptation, meaning that on-the-fly, without us doing anything, when it doesn’t succeed it’ll update its understanding of the scene and try some new things.”

    Prikaži ovu nit
    Poništi
  10. prije 22 sata

    It sounds like they trained on both sim2real and real data as well

    Prikaži ovu nit
    Poništi
  11. prije 22 sata

    The robot is able to perceive pretty challenging situations: “it is able to detect glossy, shiny, and reflective products, including products in plastic bags. The product range is nearly unlimited, and the robotic picking station has the same or better performance than humans.”

    Prikaži ovu nit
    Poništi
  12. prije 22 sata

    The robot is also working in the real world: “the system has been up and running reliably and cost effectively in a real warehouse in Germany for the last four months. “

    Prikaži ovu nit
    Poništi
  13. prije 22 sata

    “It has to be a single network able to handle any kind of [product], any kind of picking station. In terms of being able to understand what’s happening and what’s the right thing to do, that’s all unified.” They call it Covariant Brain.

    Prikaži ovu nit
    Poništi
  14. prije 22 sata

    It sounds like they have one giant neural network that generalized across customers, which is exciting: “What couples the vision system to the suction gripper is one single (and very, very large) neural network, which is what helps Covariant to be cost effective for customers.”

    Prikaži ovu nit
    Poništi
  15. prije 22 sata

    They use a pretty standard robotics hand with a suction gripper, as well as simple standard 2D cameras (no lasers, etc).

    Prikaži ovu nit
    Poništi
  16. prije 22 sata

    They focused on the picking problem: “taking products out of one box and putting them into another box”

    Prikaži ovu nit
    Poništi
  17. prije 22 sata

    Many robotics startups IMHO focus on the tech first and choose sectors that already have rock bottom labor costs or involve commodity products, which makes it very hard for smarter robots to compete. Looks like Covariant did a good job along the product dimension.

    Prikaži ovu nit
    Poništi
  18. prije 22 sata

    One thing about logistics and warehouses is it can’t just be outsourced to other countries, where robots would have trouble competing — by definition most logistics warehouses are near the customer, so robots can be competitive

    Prikaži ovu nit
    Poništi
  19. prije 22 sata

    “logistics especially is just hurting really hard for more automation. And the really hard part of logistics is what Covariant decided to tackle.”

    Prikaži ovu nit
    Poništi
  20. prije 22 sata

    They came down to manufacturing and logistics

    Prikaži ovu nit
    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·