The goal of the project described in this paper is to develop a human-adaptive Scrub Nurse Robot (SNR) that can adapt to surgeons with various levels of skill and experience in order to compensate for the present severe shortage of scrub nurses. To determine the specifications of the SNR, we analyzed real intraoperative behavior of a scrub nurse, and then modeled the entire surgical procedure with key participants by a multilevel modeling approach using the extended timed-automata-based formalism of Uppaal. Specifically, first, we videotaped the intraoperative motions of a scrub nurse and a surgeon in a thoracoscopic surgery performed on an infant pig, and analyzed their motions during the skin incision. Second, the motions of the nurse's right wrist, elbow, and shoulder were modeled with the timed automata. Third, the entire surgical procedure as well as actions and statuses of key participants was also modeled. Finally, it is shown that the proposed multilevel modeling approach also facilitates the model checking that is considered efficient in the SNR motion analysis and its adaptive motion planning.