AI Control - Reaction wheel Pendulum

We used the advanced control methods based on Intelligent control and big data analytics to develop a real-time adaptive control algorithm. Our developed controller shows superior over the state of the art state-space based controller.

[@richardbloemenkamp8532]

Sorry but this comparison means not much. There isn't even a reference to a technical implementation or concept. The state-space solution looks badly implemented and should work flawlessly for this simple example. I think we should maybe stop with superficial 'magical' A.I. demonstrations of things that we were able to do 50 years ago and do some more challenging work with more explanation.

[@londonnight937]

The state-space based controller is performing badly just because it's badly programmed. Having that aside, the AI-based solution is an interesting approach.

[@wojciechw8544]

I totally agree. I did a similar project using state space, and it worked much better than here. I was able to not only stabilize the pendulum, but also slow down the motion of the wheel. See my project: kzbin.info/www/bejne/boPWn6J3mZJ6ias

[@fanjan7527]

Absolutely fascinating

[@revimfadli4666]

Fascinating, what kind of AI and training algorithm did you use?

[@seananderson7087]

Interesting video! Would've loved to see more than a few seconds for each test though, though I'm sure others would've complained the other way around as well.

[@Jono.]

That's way cool. I wish I knew how to do stuff like this!

[@neortik7]

im curious to know what is the difference in the code. Maybe the is one use dynamic variation and the state one use flat variation ? Like : if on the left go up and if on the right go up. And the ia on is more like : if a little low go a little up if a lot low go a lot up ?

[@TheJacklwilliams]

This is fantastic and shows clear justification for choosing the AI methods. My challenge as I begin building my BOM and planning out a build is, your comment above re big data analytics. Now keep in mind my knowledge of the various machine learning techniques at this point is quite shallow. Given the special use case of the device, how were you able to gather big data to analyze? Aka, the assumptions is gigs or teraflops, what have you. What was the data source? I can't picture myself sitting with this for ten thousand run throughs to get enough data for it to "learn". And yes, just noticed, youtube kicked this towards me because of my endless searching and delving into this. It's been three years, my assumption is this project has been completed and your team has moved on to bigger and better things. Anyway, great shot, gave me a lot to think about regarding the approach and execution of this.

[@OzzMazz]

These post are nearly always teasers and don't go into enough detail. I'm trying to build one and just making sense of the IMU signals is a big enough obstacle to progress. Where can one find information on how to use the IMU values?

[@alihancoban953]

what is the al-based control or controller, do you have any documentation about these that you can recommend to me. thank you

[@3i33i]

wanna know how to do

[@lizbethyasminehuamanicondo8115]

do you have any paper? that's amazing

[@forbiddenera]

Iwantpapar

[@joeylawell3590]

Can i buy one yet

[@joeym9169]

What controller did you use for this system?

[@drumbum7999]

you mean what microcontroller? 2 different controllers were demonstrated here. A state-space based controller vs. an AI based controller.

[@charimuvilla8693]

Damn I thought PID controllers are something AI wouldn't touch. Clearly AI has a lot to add

[@richstaubin333]

First. Lol seriously guys good work.

[@sangminlee1278]

wtf is this..... holy..

[@Stinktierchen]

Since when is better and optimized code called AI? People stop calling everything AI.. thx