A Comparison of Differential Kinematics and Machine Learning Approaches in Motion Planning for Intra-operative Cardiac Ultrasound Robots

• Main Authors: Lin, Haichuan, Xie, Yiping, Hou, Xilong, Chen, Chen, Wang, Shuangyi
• Format: Conference Proceeding
• Language: English
• Subjects: Automation; continuum robots; Kinematics; machine learning; Mechatronics; modeling and simulation; Planning; Reinforcement learning; robotic ultrasound; Supervised learning; Ultrasonic imaging
• ISSN: 2152-744X
• DOI: 10.1109/ICMA57826.2023.10216182

Intra-operative cardiac ultrasound robots are emerging as a new tool to assist in cardiac interventional procedures to improve the efficiency of the procedure and reduce the operator's experience. These robots are often based on a continuum mechanism to achieve end position changes. The central issue for the control of these robots is motion planning. With its flexibility, machine learning-based approaches may provide new ideas for this type of problem compared to traditional differential kinematics, although related research is limited. In this paper, we present specific approaches to achieve the corresponding tasks by differential kinematics, supervised learning and reinforcement learning for specific motion planning problems, and thus analyze the feasibility of different approaches and compare their performance. The results show that differential kinematics is able to perform the best considering all tasks, while both supervised learning and reinforcement learning methods can also accomplish the intended task within a certain error range and with different characteristics. To conclude, the work lays the foundation for further flexibility in solving more complex tasks in motion planning using the combination of conventional kinematics and machine learning-based methods.