From conformation to exploration: advancing autonomous robotic manipulation of soft organs in laparoscopy

Robotic-assisted minimally invasive surgery (RAMIS) has transformed clinical practice by reducing patient trauma, shortening recovery times, and improving surgical precision. Yet despite the advances of teleoperated systems, current surgical robots remain entirely dependent on manual control, with autonomy limited to auxiliary tasks. To progress towards autonomous surgery, robots must not only perceive the complex environment of the abdominal cavity but also interact intelligently with deformable tissues whose shapes evolve continuously under manipulation. Achieving this requires new methods that unify perception, modelling, and planning within a framework capable of operating in real time and under clinical constraints.

This thesis investigates the fundamental operations that underpin autonomy in robotic laparoscopy, focusing on the ability to conform organs into desired shapes and to explore unknown tissues when no target configuration is specified. At the core of this work lies the challenge of representing soft tissue in a manner that is both computationally efficient and geometrically faithful, allowing robotic actions to be planned and executed reliably despite occlusions, sensor noise, and limited prior knowledge of material properties.

The first part of the research introduces a vision-driven, model-free approach for organ conformation. By employing a Grid-Point-Based Weighted Residual Method (GP-WRM), this approach directly relates observed surface deformations to robotic motion. It eliminates the need for explicit biomechanical modelling, while remaining robust to occlusions and variability across tissue types. The framework is validated extensively in simulation and through experiments on phantom and cadaveric tissues using the da Vinci Research Kit (dVRK), demonstrating reliable shape control across diverse surgical scenarios.

Building upon this, a physics-based framework is developed that integrates wavelet decomposition with the boundary element method (BEM). This enables multiscale representation of soft organ surfaces, preserving both global and local geometric fidelity while avoiding the instabilities inherent in Jacobian-based formulations. The method operates effectively under partial observability, supporting real-time control without reliance on volumetric meshing or extensive parameter identification. Experiments on phantom and ex vivo organs confirm stable, precise manipulation at millimetre accuracy, even under challenging conditions.

The final part of the thesis addresses the autonomous exploration of unknown soft objects, a task essential for procedures such as searching for hidden anatomical structures. A topology-aware reconstruction framework is presented that combines cylinder-Čech complexes (CCC) with point cloud data to build canonical models of deforming organs and detect topological changes as they occur. This representation informs a motion planner that selects grasping points and plans manipulations, such as turning-over and stretching, that systematically reveal occluded surfaces. Through simulated and physical trials on both synthetic objects and human cadaveric organs, the framework achieves comprehensive surface exploration within a small number of manipulation cycles, significantly outperforming existing point-cloud based methods.

Together, these developments provide a coherent framework for autonomous robotic manipulation of soft tissues in laparoscopy. By seamlessly integrating model-free, physics-based, and topology-aware strategies, the research demonstrates how real-time perception and control can be coupled with intelligent exploration to overcome the inherent challenges of deformable object manipulation. This work thus advances the state-of-the-art (SOTA) in surgical robotics, laying important foundations for the future of autonomous minimally invasive procedures where robots may assist surgeons not only as tools but as active partners in the operating theatre.