Project Description
Robotic colonoscopy is emerging as a promising solution to reduce the physical burden on clinicians and improve the safety and consistency of minimally invasive procedures. However, navigating a flexible
endoscope through the complex and highly deformable colon remains a challenging task. The environment contains tight bends, uncertain tissue interactions, and continuously changing visual cues,making reliable navigation difficult using traditional control strategies alone.
Recent advances in learning-based perception and control provide new opportunities for autonomous navigation in robotic endoscopy. By combining endoscopic vision, machine learning models, and control
strategies, robotic systems can learn to predict steering commands that guide the endoscope safely through the colon while minimizing excessive contact forces and unstable motions. These methods allow
the system to adapt to the highly variable anatomical environment encountered during endoscopic procedures.
In this master’s thesis project, we will investigate learning-based navigation strategies for robotic colonoscopy. The student will develop algorithms that map visual observations and/or system states
to steering commands for a flexible endoscope. These approaches will be evaluated in simulation environments and validated using experimental datasets from flexible endoscopic systems. The project
will explore how learning-based models can complement classical control techniques to improve navigation robustness in complex anatomical environments. Ultimately, this research contributes to the
development of intelligent robotic endoscopy systems capable of assisting clinicians or performing semi-autonomous navigation during minimally invasive procedures.
Objectives
- Study the principles of robotic colonoscopy, flexible endoscope mechanics, and navigation challenges
in deformable anatomical environments. - Investigate learning-based methods that predict steering commands from endoscopic visual feedback
and/or system state information. - Develop and implement a navigation model that maps observations to control signals for guiding
a flexible endoscope. - Evaluate the proposed navigation approach using simulation platforms and/or experimental
datasets from robotic endoscopy systems. - Compare the performance of learning-based navigation strategies with classical control approaches
in terms of navigation accuracy, stability, and robustness.
Requirements for students:
- Background in robotics, control systems, or machine learning
- Experience in programming (Python and/or C++)
- Familiarity with deep learning frameworks such as PyTorch or TensorFlow
- Interest in medical robotics and minimally invasive surgical systems
- Experience working in Linux environments (desirable)