ABSTRACTThis study investigates group navigation with the aid of strong interaction between two kinds of agents: A shepherd drives a sheep group with a large population to a given goal position. Even though numerous studies have been performed on the realization of shepherd-like navigation, they are based on the condition that all sheep positions are given. This study examines the navigation of a sheep group using a local-camera-based approach, i.e. a shepherd perceives sheep using the shepherd's vision. Before testing local-camera-based navigation, we design a shepherd controller referred to as a farthest-agent targeting controller, in which the shepherd selects the sheep farthest from the goal. We demonstrate the validity of the proposed controller using statistical analysis and comparison with previous conventional controllers. After examining the effectiveness of this controller, we show that the controller works appropriately even if the shepherd cannot know all sheep positions. In addition, we show the robustness of the proposed controller for the positional errors of the sheep flock or for agent-lost cases to apply it to real-world situations.