Abstract: A segmented on-line motion planning method is proposed for the free-floating space manipulator capturing moving targets, mainly to settle the LOS (line of sight) blocking problem of the stereo vision system probably caused by the manipulator, and to satisfy the time requirements for the capturing. An estimator of target motion states is built based on extended Kalman filter (EKF). The LOS blocking constraint of stereo vision system is modeled. The process of approaching the target is divided into two stages. The rolling RRT (rapidly-exploring random tree) algorithm in multi-constraint environment and the proportional navigation algorithm for approaching the target quickly are used in the two stages respectively. The switching between the two stages is performed according to the estimated accuracy of the target motion autonomously. Moreover, the angular velocity of the carrier is limited in the path planning according to the dynamic coupling of the combination. Mathematical simulations are conducted to verify the estimator of target motion and the path planning method of the manipulator. The proportional navigation algorithm may fail to capture the target due to the LOS blocking of stereo cameras caused by the manipulator, while the segmented motion planning method in this paper succeeds in capturing the target in all simulation conditions. The proposed segmented motion planning method is able to estimate effectively the motion states of the target, and capture the targets moving to different directions quickly and reliably, to avoid capturing failures caused by the LOS blocking of stereo vision system. The switching strategy based on the estimation of target motion can switch autonomously between two stages according to the actual motion of the target online, and shows robustness to the targets with unknown movement states.