Tree Structure Optimization

Feature Requests longterm
1

Hi,

I have the following problem. I am trying to solve the optimal control problem that has a tree structure. For example, I start with the current state and I make two branches from there. In the next step I have two nodes. From each of those nodes I make two new branches so that in the next stage I have four nodes and so on. I connect to each of these nodes an input variable that I want to use as an optimization variable and noise variables that are parameters in the problem. Is it possible to formulate a problem using acados where at each stage I would have different number of optimization variables and different number of parameters (with different values)? In general, can acados be used to optimize such a tree structure where I would define the problem stage by stage?

2

Hi Petar,

is the OCP nonlinear?
If it is linear, I would recommend GitHub - dkouzoup/treeQP: Tree-Sparse Quadratic Programming
Unfortunately, there are no interfaces in acados to solvers that exploit the tree structure.

This is possible from C, but not through the interfaces to Matlab/Octave & Python.

3

Hi Jonathan,

Thanks for the answers. Unfortunately, the OCP is nonlinear.

Are there any plans to make such interface in the future?

4

Also, is there a simple script or example where a similar idea is implemented in C?

5

I don’t know about such plans unfortunately.

Similar meaning?
I think there is no actual example with varying state dimension, and state transition functions for this case need to be thought of.

About different numbers and values of parameters (for cost, constraints and dynamics functions), this is possible also through Python, Matlab.

6

I thought that exactly. Thanks for the answers.