Trajectory Learning Using HMM: Towards Surgical Robotics Implementation

Area of Science:

• Robotics
• Surgical Technology
• Machine Learning

Background:

• Surgical robotics autonomy is advancing rapidly.
• Learning from Demonstration (LfD) is crucial for developing autonomous surgical systems.
• Current LfD methods require enhanced trajectory representation.

Purpose of the Study:

• To improve trajectory generalization in high-dimensional spaces for surgical robotics.
• To develop a more comprehensive representation of demonstrated trajectories using multidimensional data.
• To enhance the codification and interpretation of information for LfD in surgical applications.

Main Methods:

• Extended the Douglas-Peucker algorithm to include kinematic and dynamic trajectory data.
• Collected and preprocessed motion and force interaction data.
• Trained a hidden Markov model (HMM) comparing motion-only versus motion-and-force data.

Main Results:

• Including force interaction data improved trajectory reconstruction accuracy.
• Achieved a lower root mean squared error (RMSE) of 0.29 mm with force data, versus 0.44 mm without.
• Demonstrated enhanced generalization in high-dimensional spaces.

Conclusions:

• The proposed method effectively encodes, simplifies, and learns robotic trajectories.
• Incorporating multidimensional trajectory data, including forces, is beneficial for LfD in surgical robotics.
• This approach advances the development of autonomous surgical systems.