Learning-based intelligent trajectory planning for auto navigation of magnetic robots

Area of Science:

• Robotics
• Artificial Intelligence
• Biomedical Engineering

Background:

• Electromagnetically controlled robots offer potential for minimally invasive surgery and targeted delivery.
• Precise automatic navigation is hindered by complex environments and system delays, making traditional kinematics modeling challenging.

Purpose of the Study:

• To develop a learning-based trajectory planning strategy for autonomous magnetic robot navigation.
• To overcome the limitations of kinematics modeling in robotic control systems.

Main Methods:

• A Long Short-Term Memory (LSTM) neural network was utilized to create a global mapping between electromagnetic actuation sequences and robot trajectory coordinates.
• A dataset was generated by manually controlling the robot along a curved path 50 times for training the LSTM network.

Main Results:

• The trained LSTM network successfully generated control sequences for automatic navigation of the magnetic robot.
• The system demonstrated effective control on both familiar curved paths and novel tortuous, branched paths within simulated vascular environments.

Conclusions:

• The proposed learning-based strategy enables effective automatic navigation of magnetic robots without requiring explicit kinematics models.
• This approach shows promise for advancing the clinical application of robotic systems in medicine.