Improving PPF algorithm for workpiece grasping with adaptive pose estimation

Abstract

In order to achieve pose estimation for robotic arms in unstructured grasping scenarios, meeting the demands of unmanned and intelligent industrial production, research has been conducted on the unstructured grasping and pose estimation of robotic arms. Firstly, the scene point cloud was preprocessed, introducing an adaptive statistical filtering algorithm to address the denoising issues encountered in traditional statistical filtering. Subsequently, target extraction was performed using an improved PPF algorithm for point cloud registration. Finally, precise pose estimation was accomplished through ICP registration, and algorithm validation as well as grasping experiments were conducted on both public datasets and data collected in laboratory environments. The experimental results indicate: After conducting grasping experiments in the grasping environment of existing laboratory equipment, it is obtained that the pose recognition accuracy and grasping success rate of our algorithm reached 93. 23 % and 85. 61 %, respectively. The recognition time was 37. 21 seconds, and the total time consumed was 68. 33 seconds, meeting the requirements of the established applications. Therefore, under the rhythm conditions of industrial production, this method ensures the robustness and accuracy requirements of pose estimation, achieving satisfactory results.