custom function

[Julian Balletshofer]

Hi,

We have a code for dynamics consisting of a nominal part and a part that gets learned online via Gaussian Process Regression. We compute the nominal dynamics symbolically and for the GPR part we derived the analytical expressions and compute the Jacobian and evaluation (prediction) numerically. The dynamics is used in an Extended Kalman Filter and currently focusses on state observation only. We now want to use CasADi for solving an optimal control problem.

I wondered if it is the best way to use the callback function to provide the jacobian computed by the GPR code or if there is a more efficient way. The nominal part could be easily rewritten to align with casadi I guess.

would it also be possible to create a function and specify the Jacobian or completly neglect the symbolic.

How would a callback function lookalike? I tried to write one but I always get error with calling get Jacobean.

all best

Julian