Development and Evaluation of Sensorless Model Predictive Force Control Schemes for Single Linear Actuators in Surgical Applications

Ensuring safety in human-robot interaction is of paramount importance in surgical robotics applications, where excessive forces and velocities could cause irreversible damage to patients or handled tissues. Traditional controllers often fail to natively handle these types of constraints, thus risking tissue perforation. This thesis presents the development and evaluation of various Model Predictive Control (MPC) architectures for the force control of linear actuators, specifically designed to guarantee patient safety. A distinctive feature of the proposed architectures is the use of a Kalman Filter configurated as an Unaccessible Input Observer (UIO), acting as a state observer, which allows for a real-time estimation of the force exerted by the motor on the interacting environment. Specifically, three different approaches were analyzed: ˆ Linear MPC (L-MPC): based on a linear tissue model, represented by its average stiffness. ˆ Non-linear MPC (NL-MPC): which integrates a non-linear model of tissue stiffness. ˆ Adaptive MPC (AMPC): which utilizes a Recursive Least Squares (RLS) algorithm to perform online estimation of environment characteristics, enabling the system to autonomously adapt to different tissue types. Experimental results demonstrate that while the linear and non-linear architectures achieve good performance on modeled tissues, the AMPC provides greater exibility and high performance even in unmodeled scenarios. Notably, all three proposed architectures achieved faster convergence, lower overshoot, and fewer oscillations compared to a more conventional Proportional controller with a Disturbance Observer. Furthermore, the transition between free motion and contact was also managed by investigating an effective contact detection strategy.