An Efficient PCA-Based Target Pose Estimation Algorithm for Low-Precision Point Clouds

This paper introduces a grasping pose estimation algorithm tailored for two-fingered parallel grippers and consumer-grade depth cameras that can only capture low-precision point cloud information. Distinguishing itself from deep learning-based methods, this algorithm does not necessitate prior training on extensive datasets. Instead, it only requires acquiring a single frame of point cloud data from the target object's surface. After projection and subsequent two-dimensional PCA analysis, the algorithm adaptively generates grasping poses suitable for the current task, while considering the mechanical structure of the robotic arm. The experimental results demonstrate that the algorithm proposed in this paper achieves a high success rate in grasping common everyday objects. Furthermore, this algorithm can readily adapt to various combinations of robotic arm configurations and grippers.