Alphabet's internet Loon balloon kept on station in the sky using AI that beat human-developed control code
- Reference: 1606998610
- News link: https://www.theregister.co.uk/2020/12/03/loon_ai_balloons/
- Source link:
The 15-metre-wide balloons [1]relay internet connections between people's homes and ground stations that could be thousands of kilometres apart. To form a steady network that can route data over long distances reliably, the balloons have to stay in place, and do so all by themselves.
Loon's AI-based solution to this station-keeping problem has been described in a research paper [2]published in Nature on Wednesday, and basically it works by adjusting the balloons' altitude to catch the right wind currents to ensure they are where they need to be.
The machine-learning software, we're told, managed to successfully keep the Loon gas bags bobbing up and down in the skies above in the Pacific Ocean in an experiment that lasted 39 days. Previously, the Loon team used a non-AI controller that used a handcrafted algorithm known as StationSeeker to do the job, though decided to experiment to see whether it could find a more efficient method using machine learning.
"As far as we know, this is the world's first deployment of reinforcement learning in a production aerospace system," [3]said Loon CTO Salvatore Candido.
Facebook quietly kills its Aquila autonomous internet drone program [4]READ MORE
The AI is built out of a feed-forward neural network that learns to decide whether a balloon should fly up or go down by taking into account variables, such as wind speed, solar elevation, and how much power the equipment has left. The decision is then fed to a controller system to move the balloon in place.
By training the model in simulation, the neural network steadily improved over time using reinforcement learning as it repeated the same task over and over again under different scenarios. Loon tested the performance of StationSeeker against the reinforcement learning model in simulation.
"A trial consists of two simulated days of station-keeping at a fixed location, during which controllers receive inputs and emit commands at 3-min intervals," according to the paper. The performance was then judged by how long the balloons could stay within a 50km radius of a hypothetical ground station.
The AI algorithm scored 55.1 per cent efficiency, compared to 40.5 per cent for StationSeeker. The researchers reckon that the autonomous algorithm is near optimum performance, considering that the best theoretical models reach somewhere between 56.8 to 68.7 per cent.
When Loon and Google ran the controller in the real experiment, which involved a balloon hovering above the Pacific Ocean, they found: "Overall, the [reinforcement learning] system kept balloons in range of the desired location more often while using less power... Using less power to steer the balloon means more power is available to connect people to the internet, information, and other people."
The balloon carries a payload of about 100kg, taking with it various electronics to collect solar energy, and communications equipment. Flying a balloon to a specific location is a different problem altogether.
"A fleet dispatch system assigns each balloon to a particular location, and a navigation controller brings it in the vicinity of that location before switching control to the station-keeping controller," the paper explained.
Loon will continue testing its balloons to provide wireless broadband service to areas that lack traditional fiber optic cables. It has sent one to Kenya and plans to send another to South America over the Amazon Rainforest. ®
Get our [5]Tech Resources
[1] https://loon.com/technology/
[2] https://www.nature.com/articles/s41586-020-2939-8
[3] https://medium.com/loon-for-all/drifting-efficiently-through-the-stratosphere-using-deep-reinforcement-learning-c38723ee2e90
[4] https://www.theregister.com/2018/06/27/facebook_kills_aquila/
[5] https://whitepapers.theregister.com/
Over complicated?
OK it was only 2 axis station keeping but a certain HMS vessel was automatically station keeping to within a foot or two, in various sea states, 30 years ago.... its kind of important when you are maintaining an umbilical, amongst other things, to keep things very steady at the top.... so whats another degree of freedom between friends?
/mines the one with the 12" data analysis tape from the Station Keeper in its pocket... also a cattle prod in the other to poke the analyst with - she didn't bother checking for PEOT so I had to hand wind it out of the drive at least once.
Re: Over complicated?
The challenge is to know how to use that third degree of freedom to control the other two.
Since the loons aren't flying anywhere, they're floating, and using existing currents to move them - and they have no brakes.
Now try station keeping a sub when the only controls you have are "up" and "down"
Re: Over complicated?
An HMS vessel may have motors to keep its position. We're not talking here about a airship but a hot air balloon: moving following X and Y-axis is possible only by catching the appropriate wind pushing in the right direction. And to do so the only way is to intervene on the Z-axis.
Loon?
Fit like Loon?
Where I came from a 'loon' is an excitable young man.
The AI is all good and well
Until it decides that the problem with the world is the little vermin humans scurrying around under it.
And we know how that story ends, see icon.
Hmmmm
The control system takes a set of inputs (such as location, wind speed, solar elevation etc. and derives outputs that control the balloon. It is thusly an input to output transform. When I see this, I set the output value thus.
The world has known for a long time issues with neural networks such as over-fitting / over-training leading to a lack of generalisation and how the NN may generalise and provide an output for data for which it has never seen relevant data before.The environmental range that the system 'succeeded' in isnt clear. Its not clear when it becomes 'better' as it learns and whether it may fail disastrously before it reaches that point.
Another thing the world has known for a long time relates to comparing new shiny performance to a really shite initial baseline. We never really know how good the 'handcrafted algorithm' StationSeeker really was. As I was told decades ago, 'If you want your system to show a great improvement, start off bad'.
So its all very lovely but "this is the world's first deployment of reinforcement learning in a production aerospace system" is stretching it a bit.
Re: Hmmmm
Yeh, this looks like someone tied a few Kalman filters together and labeled it "AI".
Loons aint Crazy Insane ..... whenever Genius is the AIM in Sees of Manic Madness
that used a handcrafted algorithm known as StationSeeker to do the job,..
On first quick reading of that what I thought I saw rendered and registered was StallionSeeker rather that StationSeeker.
That algorithm can certainly retain practically anything in an appropriately fetching position for Loon Teams Remote Command and Control Situation Planning Networks ...... Advanced IntelAIgent Virtual Administration of Future Programmed Projects via AIdDivine Interventions
You may like to consider all of the above as such a one. For some it will be their first, for many is it long overdue and much welcomed and for a canny few, the gift from you know who and what of another one to utilise better than ever with all that was never ever enabled before.
That advises that these such AIdDivine Interventions are Fundamentally Unique and may even be Pure and Raw Untested. There again, they may well have been Extensively Tested Guaranteeing Perfect Future Performance in Failed/Fallen Over SCADA Systems ....... those Awash and Drowning in Distressed Assets and Expensive Loss Leading Liabilities ...... and which the Consequence of their Former Activities.
Is there a better offering available anywhere else that you know of, and which can be contacted for more information on the nature of their Product? Bettering Perfection is certainly something Captivating and who wouldn't want to sample such a Ware for Distribution/Presentation/Augmented IntelAIgent Virtual Realisations for Transfer to Enabling Bodies on Earth.
Quite who you can reasonably and realistically identify as Enabling Bodies on Earth would be helpful question to be able to answer. They may have questions of their own to ask and to know where to chat with them is Vital and a Real Boon for Teams Experienced and Experimenting in All Virtual Matters that Really Matter on Earth.
Methinks that would be One Helluva Almighty Algorithm to Engage and Entangle and Entwine with :-)
There aint no mistaking that for no gospel truth, that's for sure.
Whatever happened to the Google C.H.E.E.S.E. Program ..... [1]https://archive.google.com/jobs/lunar_job.html
[1] https://archive.google.com/jobs/lunar_job.html
Re: Loons aint Crazy Insane ..... whenever Genius is the AIM in Sees of Manic Madness
https://www.mk.ru/moscow/2020/12/03/sinoptiki-predupredili-o-prikhode-v-moskvu-marsianskoy-pogody.html
Moscow is Captured by a Martian Weather
Re: Loons aint Crazy Insane ..... whenever Genius is the AIM in Sees of Manic Madness
"I thought I saw rendered and registered was StallionSeeker rather that StationSeeker."
As we all know if there's something on the internet, "There's a porno of that" - and I'm now afraid to look
Re: Loons aint Crazy Insane ..... whenever Genius is the AIM in Sees of Manic Madness
Exactly this what you mean is NOT a porno. Sorry.
AI?
So a GPS signal, altitude monitor and steering mechanism?
I've an AI keyboard. I press a "key" marked "A" and it uses AI and ML in a blockchain to make it put a "A" on my screen
Clever stuff.
*cough* ...
"by taking into account variables, such as wind speed, solar elevation, and how much power the equipment has left. The decision is then fed to a controller system to move the balloon in place."
Sorry explain to me again where the AI is in this?
Also, just asking for a friend... but how high do these things float about? My friend *cough* is interested to see what their resistance would be to a round from a high velocity sniper rifle or lightning bolt.
Sent 1 to Kenya? A quick check on Flightradar shows they've currently got 15 of them there.