ABSTRACTDeal with the dynamic environment problem, in this paper, we propose a novel scheme contains moving object detection and environment reconstruction for the interactive robot using a pitch-and-go laser scanning system. We first propose a point clouds subtraction method to separate the point clouds data to static and dynamic parts. To reconstruct the static environments, the 6D transforms between different robot poses are first estimated by a normal distribution method and then optimized with the pose graph. Moreover, for the dynamic objects in the dynamic point clouds, we propose a mean axis descriptor to recover their motion distortions so that they can be recognized. Therefore, we contribute to the interactive robots not only the reconstructed static environments but also the recovered dynamic human targets. The experimental performances on both the public RGB-D datasets and the real dynamic scenes observed by slow-scanning LIDAR indicate that the proposed methods are accurate and efficient.