Robots are amazing things, but outside of their specific domains they are incredibly restriction. So flexibility — not physical, but mental — is a constant region of studies. A trio of new robotic setups demonstrate routes they can evolve to accommodate novel situations: using both “hands,” getting up after a autumn, and understanding visual instructions they’ve never seen before.
The robots, all developed independently, are gathered together today in a special issue of the journal Science Robotics dedicated to learning. Each shows an interesting new route in which robots can improve its relationship with the real world.
On the other hand…
First there is the question of using the right tool for a undertaking. As humen with multi-purpose grippers on the ends of our limbs, we’re pretty experienced with this. We understand from a lifetime of touching stuff that we need to use this grip to pick this up, we need to use tools for that, this will be sunlight, that heavy, and so on.
Robots, of course, have no inherent knowledge of this, which can induce things difficult; it may not understand that it can’t pick up something of a given size, shape, or texture. A new system from Berkeley roboticists acts as a rudimentary decision-making process, categorizing objects as able to be grabbed either by an ordinary pincer grip or with a suction cup grip.
A robot, wielding both simultaneously, decides on the fly( use depth-based imagery) what items to grab and with which tool; the result is extremely high reliability even on pilings of objects it’s never seen before.
It’s done with a neural network that ingested millions of data points on items, arrangings, and attempts to grab them. If you attempted to pick up a teddy bear with a suction cup and it didn’t work the first ten thousand times, would you keep on trying? This system learned to stimulate that kind of determination, and as you can imagine such a thing is potentially very important for undertakings like warehouse picking for which robots are being groomed.
Interestingly, because of the “black box” nature of complex neural networks, it’s difficult to tell what exactly Dex-Net 4.0 is actually basing its choices on, although there are some obvious preferences, explained Berkeley’s Ken Goldberg in an email.
” We can try to deduce some hunch but the two networks are inscrutable in that we can’t extract understandable’ policies ,'” he wrote.” We empirically find that smooth planar surfaces away from edges generally score well on the suction model and pairs of antipodal points generally score well for the gripper .”
Now that reliability and versatility are high, the next step is speed; Goldberg said that the team is” working on an exciting new approach” to reduce computation time for the network, to be documented , no doubt, in a future paper.
ANYmal’s new tricks
Quadrupedal robots are already flexible in that they can manage all kinds of terrain confidently, even retrieving from slips( and of course cruel kicks ). But when they fall, they fall hard . And generally speaking they don’t get up.
The way these robots have their legs configured makes it difficult to do things in anything other than an upright posture. But ANYmal, a robot developed by ETH Zurich( and which you may recall from its little trip to the sewer recently ), has a more versatile setup that devotes its legs extra degrees of freedom.
What could you do with that extra motion? All kinds of things. But it’s incredibly difficult to figure out the exact best way for the robot to move in order to maximize velocity or stability. So why not use a simulation to exam thousands of ANYmals trying different things at once, and use the results from that in the real world?
This simulation-based learning doesn’t always work, because it isn’t possible right now to accurately simulate all the physics involved. But it was able create extremely novel behaviours or streamline ones humen thought were already optimal.
At any rate that’s what the researchers did here, and not only did they arrive at a faster trot for the bot( above ), but taught it an amazing new trick: get up from a autumn. Any autumn. Watch this 😛 TAGEND
It’s extraordinary that the robot has come up with basically a single technique to get on its feet from nearly any likely fall posture, as long as it has room and the use of all its legs. Remember, people didn’t design this — the simulation and evolutionary algorithms came up with it by trying thousands of different behaviors over and over and maintaining the ones that worked.
Ikea assembly is the killer app
Let’s say you were given three bowls, with red and green balls in the center one. Then you’re devoted this on a sheet of paper 😛 TAGEND
As a human with a brain, you take this paper for instructions, and you understand that the green and red circles represent balls of those colors, and that red ones need to go to the left, while green ones go to the right.
This is one of those things where humans apply vast amounts of knowledge and intuitive appreciation without even realizing it. How did you choose to decide the circles represent the balls? Because of the shape? Then why don’t the arrows refer to “real” arrows? How do you know how far to go to the right or left? How do you know the paper even refers to these items at all? All questions you would resolve in a fraction of a second, and any of which might stump a robot.
Researchers have taken some newborn steps towards being able to connect abstract representations like the above with the real world, a task that involves a significant amount of what amounts to a kind of machine creativity or imagination.
Making the connection between a green dot on a white background in a diagram and a greenish roundish thing on a black background in the real world isn’t obvious, but the” visual cognitive computer” created by Miguel Lazaro-Gredilla and his colleagues at Vicarious AI seems to be doing pretty well at it.
It’s still very primitive, of course, but in theory it’s the same toolset that one uses to, for example, assemble a piece of Ikea furniture: look at an abstract representation, connect it to real-world objects, then manipulate those objects according to the instructions. We’re years away from that, but it wasn’t long ago that we therefore years away from a robot getting up from a fall or deciding a suction beaker or pincer would work better to pick something up.
The papers and videos demonstrating all the concepts above should be available at the Science Robotics site .
Read more: https :// techcrunch.com