Upstreaming MPCC controller?

[hidmic]

Now that the Safe Robot Learning Competition is over, we at Ekumen were wondering if you guys at UTIAS DSL would be interested in adding an MPCC controller to the set of supported controllers. We could adapt our implementation to fit the typical safe-control-gym controller structure and submit it.

Thoughts? FYI @JacopoPan @adamhall

[JacopoPan]

I do think that would be very nice and, by "typical" you also mean force control as in the main branch?
I want to know what @adamhall thinks but it might be worthwhile to work from the dev-experiment-class branch already.

[hidmic]

Friendly ping @adamhall

[adamhall]

Sorry! I'm just in vacation but will be back next week!

[adamhall]

Hey! I'm back from vacation. We can certainly take a look at mixing in the MPCC controller. It would probably be best to put it in the dev-experiment-class branch. The major difference is that the dynamics model might be pretty different now given that the inputs will be thrusts which I believe is different than what we put for the competition. I can take a look at your implementation to see what might be required. Where might I find it?

[JacopoPan]

@hidmic and the team have a copy of the repo at https://github.com/ekumenlabs/safe-control-gym created for the competition
I think now we should work through branches/PRs on this one instead

[hidmic]

That's about right @JacopoPan, and yes, we sure can make a PR to this repository, to the dev-experiment-class branch.

Regarding implementation changes, the dynamics model should be roughly the same and having full state observations and direct access to actuators may actually simplify things a bit (no need for estimators). FYI @glpuga.

[hidmic]

@adamhall friendly ping

[hidmic]

@adamhall @JacopoPan friendly ping, the last one before closing this issue internally.