Transfer learning on UR robots

• Main Author: Yu, Xiwei
• Other Authors: Cheah Chien Chern
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering; Robot control; Deep learning
• Online Access: https://hdl.handle.net/10356/176908
• Institution: Nanyang Technological University

As neural networks and deep learning develop, researchers are continually exploring the capabilities and potential of using data-driven methods to control robots. As a data-driven approach, training neural networks requires substantial data support, making the collection of sufficient data to create datasets crucial. However, in many practical applications, data collection is expensive, time-consuming, and sometimes impossible. Transfer learning can address the issue of data collection by fine-tuning and transferring already trained neural networks to new robots. This paper proposes a new data collection method that gathers unbiased data, achieving transfer learning from the UR10e to the UR5e robot. The transferred neural network was then tested on the physical UR5e robot, and experimental results of tracking control on industrial robots were presented, verifying the low data loss and effectiveness of transfer learning.