Add an illustrative example of active inference to the intelligent agent lesson about bmlip

[bertdv]

Not strictly but in that order but these it does provide a nice way of structuring the computations.

[wmkouw]

@MagnusKoudahl, we can just take the code from the BATMAN paper. It's mostly FL, right? A small part is Turing.

Burak's robot is impressive, but I have no idea what he has put in there. Thijs' code has this 'act-execute-observe' structure, which I'm not sure is suitable; we would have to explain what he means with that.

[bertdv]

The (1) act-execute-observe, (2) infer the next control/action, (3) slide forward structure of Thijs is needed for online active inference, but I do teach that idea in the class so it does make it very suitable I think.

[MagnusKoudahl]

I would be wary of using BATMAN as a first exposure. There is some weirdness with the binary feedback that means it takes a very large number of iterations to converge and does "jumps".
Buraks robot runs basically identical code and it's showing similar issues.

For a good demonstrator, my money would be on Thijs' Mountain Car. Act-Execute-Observe is not difficult to move from helper functions and into the main notebook if we want.

[MagnusKoudahl]

@bertdv the three steps in that order are not strictly necessary. BATMAN and the Generalised FE paper both run a different order

[wmkouw]

Alright. Thijs' mountain car seems like the more suited option.

Thijs' code is archived here, right?
@magnus, did you already check with Thijs whether this is up-to-date? I can ask him today.

[MagnusKoudahl]

I haven't checked with him yet. I've fiddled around with the notebooks a bit this morning and there seem to be some minor bugs - nothing we can't handle with a little bit of elbow grease though.