A tiny flying robot gets 450% faster with AI

This thing was thethered to an off-board power source and control system. Now they 'just' have to integrate all that too and off they go.

They could have told us so in the Science Daily blurb, but they didn't.

Jeroen Belleman

Yes, was wondering how they did it.

But still being able to fly in a very small space without hitting the walls is amazing! Their software could be used in bigger thing too.

I was reading CERN is taking their stuff apart and replacing the magnets with ever more powerful ones?

Do they still believe running 2 Tesla cars at supersonic speed against each other will make them understand its microchips ?

Was recently reading about all those particles' they have so far 'discovered' Lost overview at some point. Feels like crap.

Back to the desktop experiments and away with mathematical babble swamp.

The flying insect, using extensive external ancillary support, was a mere test and validation platform for much deeper analytical, scientific, and engineering principles that have been in development for quite some time. The fundamental "predictive control" with environmental disturbance perturbation was first elucidated in a 15-year old paper here:

Tube‐based robust nonlinear model predictive control DQ Mayne, EC Kerrigan, EJ Van Wyk, P Falugi International journal of robust and nonlinear control, 2011•Wiley Online Library Abstract This paper extends tube‐based model predictive control of linear systems to achieve robust control of nonlinear systems subject to additive disturbances. A central or reference trajectory is determined by solving a nominal optimal control problem. The local linear controller, employed in tube‐based robust control of linear systems, is replaced by an ancillary model predictive controller that forces the trajectories of the disturbed system to lie in a tube whose center is the reference trajectory thereby enabling robust control of uncertain nonlinear systems to be achieved. Copyright © 2011 John Wiley & Sons, Ltd.

And within the context of AI you have the general area of research of Model Predictive Control (MPC): Neural Network Surrogates for Speed Acceleration: Traditional MPC can be computationally prohibitive for fast, high-frequency systems (like robotics or aerospace) because it solves online optimization algorithms continuously. The AI Approach: Deep neural networks are trained offline using data generated by a traditional MPC controller. The Benefit: The trained network acts as an explicit control surrogate, mapping the current system state directly to the optimal control action instantly. This can accelerate control cycle speeds by 10x to 100x, sometimes reaching tens of thousands of Hertz.

All of this has been in development for quite some time.

The news release is not really about robotic mosquitoes performing ballet.

Science Daily is always amusing. Like Popular Mechanics.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

They don't say if it flew into the wall or not. They *do* say that the actual trajectory wiggles randomly and rapidly around the desired path. I'd hate to fly in an airplane doing that.

They have a freely accessible paper explaining all this in detail:

Yes, the LHC is being upgraded to become the "High Luminosity LHC". This involves increasing the intensity of the beams and tweaking the beam optics around the experiments. And much more...

A lot of magnets in the vicinity of the crossing points need to be replaced, all machines need to be adapted to handling more protons per shot, and not least, all experiments will need upgrades of the acquisition electronics to deal with the increased event rate.

Anyway, lots of stuff needed replacement anyway, because of several years of accumulated radiation damage, or simply because of having become obsolete.

The objective is to better pin down the properties of the Higgs and hopefully to discover some new physics. The LHC disappointed a lot of physicists. Apart from the expected Higgs, no notable other discoveries were made, although physicists closely involved in the LHC experiments will probably disagree with me on that point.

As for your analogy, you can't take protons apart with spanners and screwdrivers, and they don't have OBD ports.

Jeroen Belleman

Well, will it help, a bigger machine?

Was reading this today:

My view is: If you cannot do it on the desktop with those small particles then you cannot do it in machine the size of the Universe.

That goes for elementary particles as well as for 'fussion oops fusion powerless eeeh power'.

Fusion was already done by Farnsworth

It won't be bigger, just more efficient at causing collisions.

There is also a project to make a new collider that would be much bigger, a wide underground ring around the city of Geneva. It circumference would be nearly 100km! In 25 years, maybe.

We've never seen it happen, at least not spontaneously. We've seen gamma radiation produce particle pairs when it hits matter, but that's not what is meant by “popping into existence”.

Maybe we can do it in a machine with an intermediate size?

Intersting device, but not as an energy source.

Jeroen Belleman

Wow, talk about infinity!

Anyways a whole industry relies on it, pity it never has found something that helps humanity? OK 'html', but then... In Amsterdam at the accelerator we had (think it was 5% of the time) for your own projects. I used that to make some video related electronics. That was really fun. lots of help there too, many interesting projects. That was the days (about 1979) of playing Moon landing game on a Commodore Pet at work:-) The big systems there ran Unix, so I bought a book on Unix and on C programming. back then. That really helped when Linux came (SLS Linux) in 1992, took me an afternoon to get going with it on my home system.

Indeed.

The first steam engines were crazy inefficient. Ditto the first solar cells.

They got better.

John Larkin Highland Tech Glen Canyon Design Center Lunatic Fringe Electronics

CERN has done some sterling work in FOSS as well, especially their numerical analysis libraries, KiCAD, and iirc FreeCAD as well.

(My views on the utility of its main effort have been rehearsed at some length in this here venue.)

Cheers

Phil Hobbs

Inevitably, organisations like CERN end up doing things for their own use that are useful to others too. CERN's merit is that it then makes those efforts available freely to everyone. This is true for the web, for KiCad and also for scientific publications. Probably for a lot of other things too: Superconductors, vacuum technology, cryogenics, electronics, magnets, ceramics, RF technology, etc, etc.

Makes sense. After all, it's the tax payers that paid for it all.

In projects I worked on, contracts stated explicitly that companies were allowed to use CERN technologies for their own benefit, non- exclusively.

Jeroen Belleman

Yes, very nice, thank you. Would you mind leaving a bit of context in your replies? I often have no idea what you're reacting to.

Jeroen Belleman