Evaluation of ML-Based Grasping Approaches in the Field of Automated Assembly

Current trends in the manufacturing industry lead to high competitive pressure and requirements regarding process autonomy and flexibility in the production environment. Especially in assembly, automation systems are confronted with a high number of variants. Robot-based processes are a powerful tool for addressing these challenges. For this purpose, robots must be made capable of grasping a variety of diverse components, which are often provided in unknown poses. In addition to existing analytical algorithms, empirical ML-based approaches have been developed, which offer great potentials in increasing flexibility. In this paper, the functionalities and potentials of these approaches will be presented and then compared to the requirements from production processes in order to analyze the status quo of ML-based grasping. Functional gaps are identified that still need to be overcome in order to enable the technology for the use in industrial assembly.