Teaching robots what a dog knows intuitively. One application: discerning walkable surfaces.

Teaching robots what a dog knows intuitively. One application: discerning walkable surfaces.

Reading Time: 1 minute

Teaching robots what a dog knows intuitively. One application: discerning walkable surfaces.

“Our intuition for this was that dogs are really good at finding where to walk — where they’re allowed to go and where they’re not,” says Ehsani. “This is a very hard task for a computer because it requires a lot of prior knowledge.” This knowledge might be whether a surface is too steep to walk on or if it’s spiky and uncomfortable. It would be time-consuming to program a robot with all these rules, but a dog already knows them all. So by watching Kelp’s behavior, the neural network learned these rules without having to be taught them. In other words, it learned from the dog.



  1. The only thing dumber than a dog is the dog’s owner

  2. Fantastic “post” on Google Plus …

    Two of Favorite Subjects






    Thanks Gideon Rosenblatt

  3. This is expedient in the manner of business, but denies the AI the long term value and sense of consequence that comes with experiential learning. It’s like teaching children manners by shoveling them into their heads with a trowel.

    You get the superficial effect without any of the understanding.

    Mark well my words: this methodology will come back to bite us.

  4. Michael Verona I think Robots will come back to bite us if they are made to learn things thru experience for then there are chances that they might become self conscious……….and therefore making them learn superficially just enough to get our jobs done is much better. What do you say ?

  5. Exactly what I was doing. My interests are in being able to predict squirrel population behaviour. IE move them without destroying them, etc

  6. Michael Verona, I suppose. But you could also make the argument that it’s like genetic coding. Babies come into life with hundreds of millions of years worth of coding and they still get to use that coding to learn more.

  7. What a riot. I guess it is good to be creative and have an imagination.

    A useful control of the animals could move nonindigenous species from habitats occupied by lessor aggressive indigenous species.

    Michael Verona I assume having neural network, statistically based AI systems does not mean a conceptual comprehension system is not in place creating goals and feedback to the entire behaviour of the model. That’s how we do it. We react quickly while our frontal lobes figure out what is going on. (I have no clue how these models do this but back in the 70s conceptual understand was pretty cool.)

  8. Robots need not be human like they need 10 arms

  9. Coach G Moore Hello Moore how’re you doing

  10. Dann Van Hey Dann how’re you doing

  11. Gary Deen Hello Gary how’re you doing

  12. Robots are built to accomplish a give task.

  13. In the case of squirrels, a lot can be learned by processing the video prior to feeding it into the network cruncher.

    A first pass strips the raw video of content where the animals are inactive. So, the video can be reduced, except the animal is constantly listening, and in many cases watching (spotting for others while they are about).

    Information I found useful was light levels, time, location (on tree, on ground, up high, down low, hidden, out in the open), environmental stimulation (proximity of others, level of specific types of noise), proximity of other squirrels, visibility of spotters (other helpful squirrels), spotters messages, location relative to their territory and markers, presence of squirrel visual and auditory information.

    I use multiple close up and wide angle cameras.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Sign up here for the latest articles. You can opt out at any time.

Subscribe by email:

Or subscribe by RSS:

%d bloggers like this: