L-M ff average smc adjustment

[thdrmngm]

No description provided.

[Effo12345]

Some notes from the model:

• I'm very much confused by using the pulse generator as an input on your FF learning block. Are you meaning for that to be your time sync? If so, that should come from the ROS time block
• alpha on the EMA filter is still too high; you ideally shouldn't see any square waves from sliding mode switching after the filter. That being said, the algorithm does seem to converge pretty well to at least plausible values. In my testing, it helped pretty substantially with pitch error, but it converged to too small of a force for the roll, leaving it with ~6.5 degree steady state (8 deg before learning)
• The initial learning rate also seems a bit drastic, with there being sudden jumps in the requested force

• We can talk more about this during S'ups but you should only let new values into the filter when the robot has met steady state conditions
• The code itself looks a lot cleaner than where I've seen it in the past. Nice work