Improving the walking functions of hemiplegia patients after a stroke or brain injury is an important rehabilitation challenge. Recently, walking assist robots have been introduced in advanced rehabilitation facilities as a way to improve the efficiency of patient rehabilitation and restore their walking functions. Expert therapists can apply this device on different patients; however, such application mainly depends on the therapist’s tacit knowledge. Thus, it is often harder for novice therapists to apply such devices on different types of patients. Consequently, effective use of a walking assist robot has become a new patient rehabilitation skill. Taking rehabilitation as a service provided by medical doctors or therapists to their patients, this study aims to improve the quality of the rehabilitation service. In particular, the objective of this study is to abstract the rehabilitation skill of expert therapists in using a walking assist robot by applying a service science methodology known as skill education. Skill abstraction was performed by interviewing an expert therapist. From this interview, it was found that the expert therapist classified hemiplegia patients into four different classes. Using videos of patients walking, further analysis revealed the expert’s tacit knowledge, which was indicated by differences observed among these four groups in particular phases of the patients’ walking patterns. This study shows that by successfully obtaining explicit knowledge of part of a rehabilitation skill by using a walking assist robot (which until now was a tacit knowledge of experts), and then organizing the acquired explicit knowledge, even non-experts can easily reproduce the skill of experts in new patient rehabilitation.