For unit-step, ramp, parabolic inputs, we can estimate the steady-
state errors.
But how to estimate the steady-state errors, if I give an input is not
either of them (a random trajectory)?
Or is it possible to treat the input as the combination of unit step,
ramp, and parabolic
inputs and then to analyze the steady-state error?

Actually, the input trajectory is a real time command, and I want to estimate the output response error. It's no sense to do it by steady-state (time -> infinity). Is it correct to estimate errors by integrating the error function with respect to time?

Actually, the input trajectory is a real time command, and I want to estimate the output response error. It's no sense to do it by steady-state (time -> infinity). Is it correct to estimate errors by integrating the error function with respect to time?