Using a Surgical Physics Simulator as the Predictive Model in Model Predictive Path Integral Control for Soft-Tissue Manipulation

Abstract

Tissue manipulation is a fundamental and frequently performed sub-task in robotic surgery, but its automation remains challenging because soft tissue exhibits nonlinear deformation and complex contact interactions with rigid instruments. This work presents a simulator-in-the-loop, samplingbased MPC framework that uses CRESSim, a GPU-accelerated Unity-PhysX 5 surgical simulation environment, directly as a black-box predictive model inside a Model Predictive Path Integral (MPPI) controller. At each control step, MPPI samples candidate gripper command sequences, evaluates them through forward rollouts in the simulator, and updates the control sequence without requiring analytical tissue dynamics, Jacobians, or local linearizations. We evaluated the approach in planar and volumetric tissue-manipulation tasks, including targetpoint placement and tissue triangulation. The results show target-point positioning errors within the millimeter range, with representative trials reaching below 2 mm, demonstrating the feasibility of using a surgical physics simulator as the predictive model for closed-loop autonomous soft-tissue manipulation.