[ODE] Re: RC-servo modeling

[Matthew D. Hancher]

RC hobby servos are so jittery and sticky that expecting to get a very
high fidelity simulation is expecting too much.  Fortunately this is
not usually a problem.  Because their controllers are so stiff, it is
often acceptable to model them as ideal velocity-limited servos, as
long as you are not operating near the torque limits.  People have
sucessfully transferred walking gates from simulation to real robots
with RC servos using this approximation, even in a purely quasi-static
simulator.

If you *are* operating near the torque limits, then the behavior of
your servos will likely get more erratic and even appear
non-deterministic, because issues like stiction inside the gearbox
become critical.  Your best bet is still to use the simple motor
parameters as Russ suggests, but with two caveats.  First, make sure
you measure your servo's torque limits; they are almost certainly
lower than the value on the spec sheet.  (You should also be measuring
the servo's no-load velocity.)  Second, because the simulation will
not be perfectly reliable, you should run it a number of times varying
the parameters slightly each time, probably also adding some jitter to
the motor command to roughly model servo crappiness.  If your walking
gate works well for a range of parameters in the face of this noise,
then you can be confident that it will work well on the real robot.

mdh

Matt Hancher
NASA Ames Research Center
Official: mdh at email.arc.nasa.gov
Personal: mdh at media.mit.edu