Add SI-units to control structure

[krichardsson]

A first small step to implement SI-units, see #537.

Basic implementation of force/thrust using code from Mike Hamer. Still old battery compensation.

[ataffanel]

Looks good to me, it is about time we add some SI units in the chain :-).

[whoenig]

Your change is of course nice and clean, is usual!

I am mostly concerned about the design decision of exposing SI units in the first place. Given that the tendency of the platform is now to support many different frames/motors, having SI units will be pretty painful to maintain (one would essentially need to rescale the gains for each possible platform). That's why I'd suggest to only expose normalized thrust/torque and force new controllers to use that mode. Then, as long as the thrust-to-weight ratio is similar, very little tuning needs to be done. Of course, one could add SI units and later the normalized ones, but there is a risk that controllers simply will stick around with the SI units and that those will be almost impossible to use on custom frames (since domain knowledge is needed for tuning).

See also #537 (comment) (and discussion below).

[krichardsson]

I see what you mean @whoenig but I'm not sure it is a problem to support SI units in the API as well. To me the SI units API is something that can be used by anyone that finds it useful, and if you chose to use it, the controller will not be easy to use on a different platform. We are still in control of what we merge and can chose if we want to merge a contribution using the SI unit API, or not.

As far as I can see in your implementation in the power distribution of the SI units/normalized forces, there is not a lot of extra code to support both APIs. When my fast copy/paste implementation is upgraded to something better (probably inspired by your code), I assume it will not be too much extra work to maintain both.

What about adding a comment to the SI unit API describing the drawbacks of using it?

I'm of course arguing like this, partly because of self interest as I'm planning to add Mike's controller that uses SI-units. As I'm short on time, I want to minimize my work for now but I do hope to convert it to normalized forces later on :-)

[whoenig]

I think that makes sense - just be careful when you merge more controllers - at some point it will be difficult to get rid of them:-)

[jpreiss]

From browsing the current cflib commander.py and the CRTP commander docs it's not clear to me if there's any way to send a setpoint with physically meaningful thrust unit or not. Is that right?

If such a feature does not exist yet, would it be straightforward to add now that we have SI-unit power distribution?

[ataffanel]

It seems that the setpoint structure in the Crazyflie is not updated for SI units yet, the thrust in there is still unit-less.

The way to implement a setpoint packet with SI unit would be to add new packets to the generic commander. This should already be possible but the generic commander packet handler will currently have to do the conversion from SI to the set-point structure in the Crazyflie. Updating the setpoint structure first would be the cleanest but is not absolutely required to expose a setpoint packet with SI thrust.