Robotic Evaluation Framework for 6D Object Pose Estimation Accuracy

For numerous manufacturing organizations, the implementation of automated processes represents a crucial strategy for maintaining competitiveness in an increasingly globalized market. In order to achieve this, a number of technical systems have been developed in recent years with the specific purpose of providing support to the employees of such companies in the implementation or operation of the aforementioned systems. The recognition of components and their pose is becoming increasingly relevant in the context of flexible automation, for example, in order to facilitate the targeted pick-up of components by technical processes or communication between humans and machines. In recent years, numerous machine learning-based methods have been developed for the estimation of object poses that were previously unknown to the algorithms. However, the requirements of automated systems have not been the primary focus of evaluation. This paper presents a robot-based approach that can record and subsequently evaluate reproducible data under industrial conditions. The present research compares three distinct pose estimation methods, with the FoundationPose approach demonstrating the most accurate results.