A high-level supervisor, employing switching and logic, is proposed to orchestrate the switching between a family of candidate controllers into feedback with an imprecisely modeled process so as to stabilize it. Each of the candidate controllers is required to integral-input-to-state stabilize one particular admissible process model, with respect to a suitably defined disturbance input. The controller selection is made by (i) continuously comparing in real time suitably defined "normed" output estimation errors or "performance signals" and (ii) placing in the feedback-loop, from time to time, that candidate controller whose corresponding performance signal is the smallest. The use of integral-input-to-state stability in the context of supervisory control of nonlinear systems, allowed us to weaken the requirements on the candidate controllers being used. It also seems quite natural when the performance signals are defined as "integral norms" of the output estimation errors.