Biomechanics-Based User-Adaptive Variable Impedance Control for Enhanced Physical Human-Robot Interaction Using Bayesian Optimization

This paper presents a biomechanics-based, user-adaptive variable impedance controller designed to enhance the performance of coupled human-robot systems during motion. The controller integrates the biomechanical characteristics of human limbs and dynamically adjusts the robotic impedance parameters-specifically damping, stiffness, and equilibrium trajectory-based on real-time estimations of the user's intent and direction of motion. The primary goal is to minimize the energy expenditure of the coupled human-robot system while maintaining system passivity. To address uncertainties in human behavior and noisy observations, the controller employs Bayesian optimization combined with a Gaussian process. To validate the proposed approach, human experiments are conducted using a standard robotic arm manipulator. The results demonstrate that the controller eliminates the need for manual parameter tuning, a process that is typically time-consuming. A comparative analysis against two variable impedance controllers without user-adaptive parameter adjustments reveal significant benefits, with the controller improving combined performance metrics-such as accuracy, speed, user effort, and smoothness-by over 13%. Notably, all participants in the study preferred the optimized controller over the alternatives. These findings highlight the effectiveness of the biomechanics-based, user-adaptive variable impedance control approach and its potential to enhance physical human-robot interaction in various applications that involve repetitive or continuous motion. This article introduces a biomechanics-based, user-adaptive variable impedance controller aimed at enhancing the overall performance of coupled human-robot systems during motion. The improvements span accuracy, speed, user effort, smoothness, and user preference during physical human-robot interaction (pHRI). To address the uncertainties of human behavior and noisy observations, the controller employs Bayesian optimization with a Gaussian process.image (c) 2024 WILEY-VCH GmbH