Abstract: The underactuated robot has difficulties in realizing multiple pose variations as fully articulated robots, therefore the classical hand-eye calibration method can not be directly applied. To solve this problem, a mathematical model is developed and a hand-eye calibration method is proposed based on plane constraints. In this method, the motion of the end effector of the underactuated robot is constrained on a predetermined plane in 3D space. The pose matrices and translation vectors in hand-eye calibration are decoupled through pure translation and rotation motions of the end effector according to corresponding calibration images. Monte Carlo simulations demonstrate the effectiveness and accuracy of the method, as well as its robustness against noises. The experiment on the developed underactuated climbing robot is performed, and the localization error is controlled under 1.5 mm. The results prove that the proposed method is simple and accurate enough for practical applications.