Autonomous Suturing in Robotic Surgery Using Reinforcement Learning and 3D Visual Feedback

摘要

Autonomous suturing is a critical advancement in robotic-assisted surgery, offering the potential to enhance surgical precision, reduce workload, and improve patient outcomes. Traditional robotic-assisted suturing relies on human teleoperation, which can introduce variability and fatigue-related errors. This paper explores the integration of reinforcement learning (RL) and 3D visual feedback to develop a fully autonomous robotic suturing system. The proposed framework consists of a robotic arm equipped with a needle driver, a 3D stereo vision system for real-time depth perception, and a deep RL model optimized for suturing tasks. Our approach involves training a reinforcement learning agent in a simulated environment, where it learns optimal suturing strategies based on trial-and-error interactions. The RL model considers needle trajectory, suture tension, and tissue deformation while maximizing accuracy and minimizing tissue damage. The 3D vision module provides high-resolution depth maps to guide the robot in real time, enabling precise needle insertion and suture placement. The system is validated on synthetic tissue models, demonstrating superior performance in terms of precision, suture uniformity, and adaptability to tissue variations. Experimental results indicate that our RL-based approach outperforms traditional teleoperated suturing by achieving higher accuracy and reducing variability. Despite challenges such as real-time computation constraints and dynamic tissue behavior, this research highlights the feasibility of autonomous robotic suturing. Future improvements will focus on enhancing real-time adaptability, optimizing computational efficiency, and expanding the system’s applicability to various surgical procedures. This study represents a significant step toward fully autonomous robotic surgery.