The SheepIT project aims at developing a solution for monitoring and controlling grazing sheep in vineyards and similar cultures. The system should operate autonomously and guarantee that sheep only feed from infestant weeds, leaving untouched the vines and their fruits. Moreover, the system should also collect data about sheep activity for logging and analysis purposes. This paper presents the overall system's architecture and its rationale, with focus on the posture monitoring and control subsystem. It includes practical results, obtained from a use case. These results are encouraging, showing that the developed system is able to estimate the sheep's posture with a high accuracy, that the stimuli are applied efficiently and that sheep have sufficient cognitive capacity to learn quickly which behaviours they should avoid. Despite being preliminary, these results provide good indications regarding the practicableness of the system.