Master thesis : Benchmarking intelligent robotic grasping methods

Abstract

Robotic grasping is increasingly attracting the attention of the industry. Research about intelligent robotic grasping and computer vision are constantly evolving and improving in order to meet this demand on real applications. Benchmarking and comparison of these solutions is an essential step to further improve intelligent robotic grasping methods. This process allows us to benchmark ourselves against others, and to continuously improve our systems. This paper discusses the importance of benchmarking in the development of intelligent robotic grasping methods. By comparing the average accuracy and speed performances of different grasping solutions in real-life setups, we can gain a deeper understanding of their capabilities, limitations, and areas for improvement. The paper presents a comprehensive benchmarking study on two specific methods, AnyGrasp and IntegrIA, in both mildly cluttered environments and more challenging cluttered. This benchmark was achieved by focusing on their performance against a reference human subject, in order to get a clearer picture of progress against the most efficient and stable agent at to this day. The results of the study provide valuable insights into the strengths and weaknesses of each method and offer guidance for future research endeavors.