> > PID systems can be unstable or conditionally-stable on such systems
> > and changing system behavior can change controller response.
>
> A well modeled control system will be unconditionally stable.
> Conditionally stability only occurs when the model is
> either designed to be so, such as an oscillator, which has BIBO
> stability, yet it has a changing output, or if the designer chooses
> to minimize but not directly compensate for the effects of
> disturbance inputs unrelated to the control input. Such models may be
> conditionally stable; This is often done for economy, however, and
> not for the lack of ability.

How do PID-based systems deal with unknown changing characteristics?  For
example, suppose a robot is supposed to pick up and manipulate objects of
varying mass, center of gravity, and possibly "inertial oddness" (e.g. a
closed vessel half-filled with liquid).  From what I understand of PID
systems, they usually require that the system be tuned for a particular
amount of intertia, while some circumstances (like the above) require that
the system work with highly variable (and unknown) inertial effects.