SQP fails in racecar example

User Questions
1

Hi,

I just noticed that the racecar example only works with the RTI scheme but not with a full SQP iteration. Are there further parameters required to set? Thanks!

All the best!

2

Hi,

yes, there are some known convergence issues in this race car problem.
Some changes to make this problem work with SQP can be found here.
I was not able to fix this entirely, so sometimes, the simulation just continues with the solution obtained after maximum number of iterations is reached.

Cheers,
Jonathan

1 Like
3

Hi Jonathan,

Thanks a lot for this fast reply! In case of convergence issues it would be nice to have debug functionalities. (like the one from casadi opti class - show infeasibilities) is there something planned for acados?

All the best!

4

Hi @fach23,

You can get the full horizon of the states and controls variables from the solver, so you can “debug” quite well.
It is explained in this post: Full horizon and cost function value from S-function - #11 by AlexWin

You can get the Slack Variable values as well, I did not try this out though:

Maybe this helps you a bit!
Greets, Alex

1 Like