## Comment Re:Non-linear control (Score 1) 51

I'm not talking about discrete components; but why do I get the feeling you're not really interested?

Keep on crawling turtle man, keep on crawling.

You are looking at this from the wrong end (and I'm not disagreeing with anything you have written). We don't have the mathematical tools to model non-linear analog systems (like the 'puppy' robot in the article), but this is not stopping people from using them. In the article they describe a new approach from a mathematical modelling perspective. These robots are exploring completely nonlinear systems with infinite state space, and getting some pretty amazing results. It's just that until recently it's been more or less heuristic.

Well fair enough, I could have been clearer, but I was hoping to pitch this to people that had a basic knowledge of control theory. On reflection, this is probably too small a group. So to explain; the situation is actually somewhat reversed to the impression you have come away with. In the continuous domain (I.e. analog feedback) our mathematical models as used are only for linear systems. So if you have a non-linear system or response, you find a linear portion and stay within that range. But non-linear is extremely important even in simple systems, for example a motor saturating. You can of course *create* a non-linear continuous feedback system, but you can't use Laplace to help you model it. In the discrete domain (digital feedback), the mathematics become very simple and although non-linearities still pose a challenge, their are many more tools in the chest for modelling these transitions. But in the world of engineering nobody bothers with even that, they just buy a PID controller and tinker with the three values. So what we end up with is a discrete controller over a strictly linear system. And you can see the appeal, the maths and modelling is extremely simple, and most people in the domain know how to do it. What has been happening in the last couple of decades is that miniaturisation of electronics is starting to make analog relevant again on the one hand (due to size, power efficiency, speed of response, and the fact that noise is less of an issue in these applications due to size), and on the other you are seeing the exploration of the "soft" and inherently non-linear properties of biological systems to perform the function of extremely complex control systems within the robotics arena. The problem has been that we have had no mathematical way of modelling this, but this article is describing a new approach. So yes, perhaps I got a little excited and left out a bunch of detail, but that's because there is just so much detail required. I just can't understand why people aren't more interested in this stuff! I want to talk about the implications, not all the boring background!

No I meant the article at the top of this page. It's all interrelated stuff, but the article here is all about soft robots (using the natural properties of the materials to perform complex control function), while Chris is all about analog electronics. Chris has several YouTube videos that are well worth a watch, and tons of published papers that are worth a read if you have access to the academic web.

i appear to need to correct a misconception, I'm not referring to the original analogue phones, but the original digital variants. In addition, the first *fully embedded* cochlear implant is possible due to the work of Chris Toumazou and his team in the analog realm. Please show me a DSP you can imbed in someone's head.

Chris Toumazou, say it with me. Read the article. It is all about replacing umpteen different actuators and sensors and ludicrously complex control feedback with some springs and soft bits whose dynamic properties achieve the same function. Analog (mechanical). That's what "puppy"' is, extremely simple from the traditional control side of things, but the dynamic properties of the the materials mean that it affects a quadruped gate. All the tricky stuff is simply a by product of the material composition. How is that not awesome? Here we are using the analog properties of "soft materials" (with or without feedback) to perform the extraordinarily complex control of these robots. Analog man, it's awesome sauce.

Also, cheaper? Cheaper to change, yes, but tell me again how a million transistors to do something an analog circuit can achieve with a handful is cheaper again?

Where did I suggest that linear control and digital control were one and the same? Digital has a range of benefits, but "power consumption" ain't one of em. You take a transistor which follows a beautiful exponential curve like much of the real world (especially biology) and crush all of that information into a 1 and a 0. We are not talking small power improvements by modelling these things in analog, we are talking several orders of magnitude improvements. And it's not just power consumption, how about response time? Digital makes it easy to change my algorithms but the fastest processor in the world is still infinitely slow compared to the instant feedback available in the analog world. Seriously, look up this Chris bloke, it'll change your life.

Thank you for the first link, it is highly informative. But I am reaching, reaching, and failing to see how *prion folding has anything to do with this whatsoever*. Seriously dude, you might want to look into your medication. Not everything is related to prions...

Shhh, shhh, go back to sleep now...

What the heck has this got to do with the article? The article is not concerned with the piss easy to model motor driven by "pulse width modulation" or anything so mundane. See that picture at the top? That is a balloon picking up a cup. A balloon. Model that with your little PID controller. This is talking about replacing all your silly actuators with their low degrees of freedom, and their pitiful centralised digital feedback controller, with a batshit insane pile networked springs (from a modelling view). The exact dynamics of your motor response look a little irrelevant in context.

Well yes, but to be fair they are very *thin* bricks now.

This article is fascinating and also a bit surprising. Surprising that the engineering world is still trying to hang onto simplified digital linear control. The real world is non-linear and analog! Linear control makes things simple mathematically and deterministic, but it also extremely limiting. There is a reason that the natural world works in a fundamentally non-linear analog fashion, and that is because it's better. Want to know why mobile phones aren't the size of bricks anymore? It's because Chris Toumazou replaced all that clunky digital radio with vastly smaller more efficient analog circuits. It's also why deaf kids can get a fully embedded cochlea implant and not have to carry around a car battery. Digital is so last century people, it's time to embrace the analog renaissance!

At that moment in time Google was better than AltaVista as AltaVista was being gamed as were all the other search engines of its type. However, it was still less useful than what AltaVista had been able to provide before the bad people came and ruined it for everyone. At least if what you were looking for was a little obscure. In that instance you could use smart search terms that would find a combination of words that would find you pages that were relevant to a topic you were researching. At the time I had gotten very good at this, but the entire approach was hacked and rendered useless by the time Google had come along. This is still the case today unfortunately, and I find myself having to search forums rather than the web to find information of interest on more obscure topics. Searching through journal databases and online libraries etc can still be usefully grepped in this way, so the skill is not entirely useless.

FORTUNE'S FUN FACTS TO KNOW AND TELL: A firefly is not a fly, but a beetle.