Learning-Based Fiber Sensing for Continuum Manipulators
Machine learning on distributed optical-fiber strain data to tell where a surgical manipulator is being touched, and how hard.
What I did
- Built an ML workflow on distributed OFDR strain data for continuum manipulators, combining ROS/Vicon data collection with real-time shape-reconstruction pipelines.
- Co-developed a cascade learning system using Gradient Boosting for contact detection and a CNN-FiLM model for contact localization and force estimation in constrained environments.
- Evaluated the sensing pipeline on real robot experiments using ODiSI strain streams and calibration-based curvature reconstruction.
Built with
- ROS
- Vicon
- Gradient Boosting
- CNN-FiLM
- OFDR
- ODiSI