A Learning-Based Approach for Contact Detection, Localization, and Force Estimation of Continuum Manipulators With Integrated OFDR Optical Fiber

A team of researchers, including Mobina Tavangarifard and Jonathan S. Kacines, has published a novel method for giving flexible surgical robots a crucial sense of touch. Their paper, "A Learning-Based Approach for Contact Detection, Localization, and Force Estimation of Continuum Manipulators With Integrated OFDR Optical Fiber," tackles a major challenge in robotic surgery. Continuum manipulators (CMs)—snake-like robots used in minimally invasive procedures—are highly dexterous but lack traditional force sensors. Their compliant structure makes it difficult to know where and how hard they are pressing against tissue, which poses a safety risk.

The team's solution is a Cascade Learning Framework (CLF) that requires only a single Optical Frequency Domain Reflectometry (OFDR) optical fiber embedded along the robot's length. This fiber acts as a dense, distributed strain sensor, capturing minute deformations caused by external contact. The AI framework first uses a Gradient Boosting classifier to detect if contact has occurred. If it has, a more complex CNN-FiLM (Convolutional Neural Network with Feature-wise Linear Modulation) model takes over to predict the exact arc-length location of the contact and the magnitude of the interaction force, outputting a spatial force distribution.

Experimental validation on a tendon-driven continuum manipulator in an obstructed environment proved the system's effectiveness. The research demonstrates that a single, integrated sensor can provide sufficient data for a machine learning model to solve the complex, coupled problems of detection, localization, and estimation simultaneously. This integrated approach is more elegant and potentially more reliable than previous methods that relied on multiple discrete sensors or complex external sensing systems.