Abstract:In human-robot interaction process, to ensure the safety of human, robots and environment in a dynamic non-structured environment, this paper presents an environment modelling method for the manipulator robot using offline-mapping between the depth image and OctoMap. This method includes a two-stage procedure: offline mapping and online updating. In the first stage, we discretize the workspace of the robot into cells and save it as a OctoMap, then get and save the offline-mapping between pixels of depth image and cells of OctoMap using intrinsic and extrinsic matrixes of the RGB-D camera. In the second stage, the OctoMap is updated by the adjacent frame substraction and lookup to the offline-mapping. Furthermore, to fully describe the robot's workspace, data fusion of multi-RGB-D cameras and robot-body-filtering based on this mehtod are also proposed. The proposed method has the advantages of powerful description ability, consistent and high accuracy, multi-sensor data fusion ability and easily online updating. Finally, the validity of this approach is demonstrated using simulations, modelling and the obstacle avoidance experiment for the flange, which is conducted by using a 7-DOF KUKA IIWA manipulator and the Kinect sensor.