# Inverted pendulum controllers needed

• posted

Hi, I'm working on an AI project, and one of the problems we're trying to get our system to solve is a virtual version of the classic inverted pendulum. We'd like to benchmark the performance of our system against that of controllers generated from standard control theory algorithms. However, we are newbies to control theory, so we need an expert to derive the controllers for us, in return for which we are willing to pay industry-standard consulting fees.

If this interests you, contact information and a description of what we need is at

.

Thanks, Francisco

• posted

snipped-for-privacy@yahoo.com (Ayala) wrote in news: snipped-for-privacy@posting.google.com:

Maybe I can save you a few dollars--

The inverted pendulum is a demo in the simulink library of matlab. With a tad of work, you can see the control equations all layed out in control systems format

Scott

• posted

Thankyou-- however, the Simulink package that I've seen (the one written by Khalil Sultan) only balances the pole, and does not also center the cart. I'm looking for controllers that accomplish both objectives.

Cheers, Francisco

• posted

snipped-for-privacy@yahoo.com (Ayala) wrote in news: snipped-for-privacy@posting.google.com:

It centers the cart if the input is set to center the cart. The demo does not have an input for external disturbances, but that's the only difference. The control equations are the same.

Scott

• posted

Your model is very idealized, yet you are asking for control methods that are used for less-than-ideal plants. What are the constraints that you are placing on yourself that you need them?

If you are assuming that the equations of motion are exactly as given (i.e. no friction or slop in the joints), that the parameters will never change, that there are no limitations on the control input, that there are no disturbances and that the pendulum will always start with the angle confined within some prescribed range (+/- 45 degrees would make it easy) then just about any controller using feedback linearization should do just fine.

In what ways do you want to vary your model from the ideal situation stated above?

• posted

I should clarify that our inverted pendulum system is virtual; there is no physical plant, only a computer simulation. Consequently, we needn't worry about friction or any other parameter that isn't explicit in the idealized model.

However, the issue of disturbances brings up one of the primary motivations of our project. According to my newbie understanding of control theory, there is usually a trade-off between performance and robustness. Our goal is to design an AI system that automatically recognizes and adapts to changing problem conditions. Using the inverted pendulum as a test problem, we'd like to see whether our system can recover its performance if we suddenly change, for example, the mass of the load on the pendulum, the magnitude and direction of the gravity vector, etc.

Francisco

• posted

Dear Francisco, It is nice hearing about your work. I did some work on Inverted Pendulm Probelm with Fuzzy Logic and as well as Neural network stuff.

I tried with H- Optimal Control analysis. I feel i got some better optimal solution when compared to Mr Kirk E Donald, famous author for Optimal control.