Deep reinforcement learning in continuous action space for autonomous robotic surgery.

Area of Science:

• Robotics
• Artificial Intelligence
• Surgical Technology

Background:

• Reinforcement learning (RL) shows promise for automating surgical sub-tasks.
• Advancements in RL enable surgical robots for complex, high-risk procedures.
• Robotic surgery aims to reduce surgeon stress and improve precision.

Purpose of the Study:

• To develop a deep reinforcement learning (DRL) approach for robotic gripper arm control.
• To optimize soft tissue cutting in a continuous action space.
• To enhance surgical pattern cutting accuracy.

Main Methods:

• Utilized DRL to control a gripper arm in a continuous action space.
• Employed a grid observation space for surgical pattern cutting.
• Developed an optimal tensioning policy to increase cutting accuracy.

Main Results:

• The DRL method demonstrated superior performance in cutting complex patterns.
• Outperformed methods using discrete action spaces and partial/random observation representations.
• Achieved higher accuracy in soft tissue pattern cutting simulations.

Conclusions:

• Introduced a DRL-based method for optimal tensioning policy in continuous action space.
• The proposed approach significantly improves accuracy in soft pattern cutting.
• This method advances robotic surgery by enhancing precision in delicate tasks.