In certain hazardous environments, teleoperation allows human operators to maintain a safe distance while controlling robots to perform tasks. Point clouds can be utilized to create an immersive environment, enhancing the accuracy and safety of remote operations. To address the inefficiency of the current 3D point cloud compression methods in robot teleoperation systems, by integrating teleoperation devices, this paper proposes a variable parameter level of detail (LoD) model point cloud compression method based on attention mechanism. By extracting the operator’s attention information, remote robots are guided to compress the point cloud, significantly improving the efficiency of point cloud compression. Firstly, a LoD compression model is designed. During runtime, operator’s eye gaze and arm stiffness information is obtained using virtual reality (VR) glasses and MYO armband, which is translated into focal point, focus range and precision requirements, dynamically adjusting the parameters of the LoD compression model accordingly. Finally, point cloud downsampling is conducted based on the LoD compression model, and serialization is performed using octrees. Experimental results demonstrate the effective improvement in point cloud compression efficiency achieved by the proposed method.