Learning Surgical Robot Autonomy from Physics Up on the da Vinci Surgical System
The presentation sketches a three-step path for raising the da Vinci Research Kit (dVRK) towards autonomous surgery.
1. Physics-consistent control. It first derives and identifies a full kinematic-dynamic model for the new dVRK-Si manipulators, unlocking accurate gravity compensation and hybrid model/learning force-estimation needed for precise, responsive motion .
2. Skill learning in simulation. A high-fidelity suturing environment (AMBF with MRI-scanned phantoms and detailed CAD tools) is built to collect human tele-operation data and train a hands-off Learning-from-Demonstration pipeline that reproduces single-throw sutures comparable or even better than experts .
3. Data-driven perception. Finally, a synchronized multi-modal dataset (HD endoscope video, kinematics, articulated key-points) is gathered on the real robot, enabling deep models for tool-tissue contact detection and bridging the sim-to-real gap .
Together, these chapters fuse accurate physics, scalable simulation, and richly annotated real data to advance surgical-robot autonomy from mere teleoperation toward conditional autonomy in the operating room.
Advisor: Professor Gregory S. Fischer (WPI)
Committee: Professor Loris Fichera (WPI), Professor Liaohai (Leo) Chen (WPI), and Professor Peter Kazanzides (Johns Hopkins University)
Zoom Link: Zoom meeting has ended.