Artificial Retinas Can Balance a Pencil On Its End

mikejuk writes "A team of researchers has built a neural information system that is good enough and fast enough to balance a pencil in real time. If you think it's an easy task, try it! The Institute of Neuroinformatics, ETH / University Zurich have used what look like video cameras to do the job but in fact they are analog silicon retinas. They work so fast that even with fairly basic hardware they can balance a pencil."

3D vision

[Anonymous Coward]

Unless I'm mistaken, from looking at the picture the camera's "eyes" are placed orthogonally, instead of side by side like a human's. That's an advantage, since we know the machine then has real 3D position info, as opposed to a human's stereoscopic 3D vision. Try it yourself: when you balance a pencil, do you fail more often sideways or towards and away from yourself?

This is an impressive bit of controls engineering, but let's not compare apples to oranges: the machine is designed for this task, and the human is not. It's in a way impressive that humans are as good as we are at this task despite not having been constructed to do it.

Re:High-Speed Robot Hand Demonstrates Dexterity

[Samantha Wright]

Job automation has been a major concern for politicians and workers since the industrial revolution, with a noteworthy resurgence in manufacturing with the introduction of automobile-building robots. Humans can out-do machines when they're underpaid (see China) but the choice to invest in a machine and all of its highly-skilled repair labour comes in response to rising wages—and by the way, did you know you're grossly off-topic?

Re:Amazing

[icebraining]

That's not really the point, is it? None of those were accomplished using machine vision.

Re:biomimetic for purely philosophical reasons

[Anonymous Coward]

In case anyone misinterprets your comment.. The fact that it is incompatible with the last decade of computer vision doesn't make it wrong, nor does it make the previous decade of research in computer vision wrong. As you wrote, different philosophies behind the solutions.

Re:To the person who told me before here on /.

[pz]

... that balancing such an object requires the use of several fancy algorithms:

This is proof that, just exactly as I asserted, all you need is relatively simple feedback as long as it's fast enough.

Yep, that was me. I guess I should go back to my MIT professors and let them all know that they're full of hooey. I've sure been shown up by Jane Q. Public!

Or, on the other hand, I could look at the video these fellows provided. Doing that, I might notice that the system is barely stable, very noisy, does not deal with perturbation very well, and accumulates error. I could then read the paper and see, under the section called "VI Control System" it explicitly states that they are using a PD system (proportional and derivative), as described in the system of two differential equations. Then I could read the sentence, "Our system normally balances an object for several minutes before losing it..." which would probably be because they don't have an I term to worry about accumulating error. Lack of an I term makes the system drift, and you can see in the video that it nearly hits the edge of the actuator workspace a few times. Striking the limit of motion would be a catastrophic change in actuator impedance and cause the pencil to be dropped. The fact that they had to include a D term means that there is more than just straight (linear) feedback. But, hey, I guess those MIT professors didn't actually know what they were talking about when they taught us 18.03 Differential Equations. Either that or Ms. Public can't read papers very well, and doesn't recognize a differential equation when she sees one.

Again, I'll state, Ms. Public, please stay away from designing any systems that are critical to support or protection of human life. You have now repeatedly demonstrated your incompetence to do so in a public forum.

Re:Samsung's automated sentry machine gun...

[Paul Fernhout]

:-)

I'm glad someone is paying attention. :-)

I was just watching some Star Trek: Enterprise episodes and when I saw all these big starships (Andorian vs. Vulcan in that case) shooting at each other, it just seemed, well, here they have warp drive, anti-matter energy, anti-gravity, and all they can think to do with it all is fight over some planets, when the whole universe is full of matter they can use to reorganize into space habitats and starships?

Anyway, I'm not saying you don't make a good point. But this is a deep issue that seems to me is being widely ignored. It relates to so many of these issue coming up as we approach one or more singularities. How do you suggest I approach that?

Re:To the person who told me before here on /.

[SomeKDEUser]

the PID algorithm is four lines of code. the RST also. But to prove the properties of either, you must understand Z-transforms (which really are Laplace transforms for sampled functions).

You can tune your PID using Ziegler-Nicholls, and that requires absolutely no knowledge of maths. To tune it optimally, you need a very good physical model, and pretty involved maths.

So I don't know what the other guy's argument were, but you might have been both perfectly right.

Re:Tag correction

[2.7182]

People have been doing this since the 90s. Here is a paper where they say they use a 30 fps camera. I am sure you can find an older one. I saw one in 1996. http://www.manuelstuflesser.net/stuflesser_paper.pdf [manuelstuflesser.net] Also, if you ask CV people they don't think they are part of AI. Some of them use AI, but there are many tools used.