Hybrid Geometric and Latent Shape Servoing for Autonomous Gallbladder Exposure
Year: 2026
Type: conference-paper
Abstract: Autonomous manipulation of deformable tissue remains challenging in robotic surgery due to complex soft-tissue dynamics and limited observability. This paper presents a hybrid visual servoing framework for autonomous gallbladder exposure in Robotic-Assisted Minimally Invasive Surgery (RAMIS). To overcome the limitations of purely geometric controllers, which can induce unsafe null-space deformations, we propose a composite state representation that combines an explicit geometric objective with an implicit physical regularizer. The liver ridge is modeled as a Bspline curve to define the exposure task, while global tissue deformation is encoded by a latent descriptor learned using a Partial-to-Complete PointNet autoencoder (AE) trained on preoperative anatomical data. A composite Jacobian-based control law jointly minimizes ridge alignment error and latent deviation, enabling safe and robust control under partial observability. Experiments on the da Vinci Research Kit (dVRK) with realistic phantoms demonstrate improved safety and robustness over standard geometric servoing.
Authors: Junlei Hu, Dominic Jones, Pietro Valdastri
Institution: University of Leeds
Language: English
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7213435467