Medical interpreters are crucial in facilitating communicationbetween healthcare stakeholders who speak different languages. Bodyorientedgestures convey critical information essential for accurate andhigh-quality interpretation in healthcare settings. This study introducesa body-oriented gesture generation system designed for medical interpreterrobots based on reinforcement learning from human feedback(RLHF). The system allows robots to interpret more naturally and improveover time through interactions with humans. By adopting a humancentereddevelopment approach, we tailor our system to address the actualneeds of healthcare stakeholders.