The International Aerial Robotics Competition (IARC) aims to move the state-of-the-art in aerial robotics forward through mission challenges. In the IARC Mission 7, the aerial robot will navigate without external navigation aids, interact with autonomous ground robots, and avoid dynamic obstacles in the herding problem. In 2017 competition, our team firstly accomplished to interactively herd one iRobot to one end of the arena through accurate and rapid aerial touch and won the first place in system control. This paper presents the self-localization, control and tracking strategies for a micro unmanned aerial vehicle (UAV) to accurately and agilely touch ground moving robots. A real-time self-localization framework is proposed to estimate the ego motion of the UAV. A computationally efficient visual tracking scheme is designed to detect the ground robot and estimate its direction. Inspired by the dance of the dragonfly, tracking and control schemes are presented for the UAV to achieve 3-D agile interaction with ground autonomous robots. Simulation and flight experimental results validate the effectiveness of the proposed methods.