Passivity-Constrained Model Predictive Variable Admittance Control for Safe and Adaptive Physical Human–Robot Interaction

Abstract

Physical Human–Robot Interaction (pHRI) occurs in industrial, medical, domestic, search and rescue, and other domains, requiring controllers that balance safety and performance for effective collaboration. In this context, this paper presents a real-time novel Model Predictive Variable Admittance (MPVA) control framework for mixed assistive and resistive scenarios, combining Variable Admittance Control with Model Predictive Control (MPC). The MPC optimizes admittance parameters based on tracking error, interaction force magnitude, and direction, while embedding passivity constraints to ensure safety and intent-aware behavior. The architecture is implemented on a 7-DoF Kinova Jaco-2 robot and validated experimentally with seven human subjects. Results, supported by objective metrics and a NASA TLX survey, show that the MPVA with force effect and passivity (MPVA-TFP) achieves competitive tracking accuracy with reduced physical effort and minimal passivity violations compared to fixed-gain and other adaptive admittance methods, demonstrating safe and effective multi-mode interaction.