Model predictive variable impedance control towards safe robotic interaction in unknown disturbance-rich environments

Abstract

Robotic systems have evolved to handle various significant interaction tasks in different environments. Under these conditions, the involvement of humans in the environment drastically complicates such interaction tasks; as the safety of humans should be prioritized while seeking to achieve the desired task aim. It is thus paramount that appropriate developments should be pursued with specific considerations for such safety-performance-balanced interaction tasks on unknown soft environments (e.g., humans). Towards this end, we present an adaptive robust, and passive control scheme based on model predictive control and variable impedance control that addresses this challenge. Under this control scheme, during robotic interaction tasks with complex environments (e.g., humans), the presented development and design incorporate safety thresholds that are carefully satisfied via impedance adaptation, and realized by a safety-related mode-switching mechanism. Once the safety thresholds are satisfied, task performance is then focused on. Additionally, a real-time adaptive robust parameter estimator is designed and utilized to estimate the environment contact model for the model predictive control, and thus this control scheme is robust against disturbances (e.g., which would invariably arise from the inevitable small bounded human motions) during the interaction tasks. Finally, the key safety and performance attainments of the proposed control scheme are verified via experiments. The experiments are conducted on two silicone rubber models and a human arm. These show that the proposed control scheme effectively outperformed various currently available control schemes in these interaction tasks with unknown environment contact models, and bounded but unpredictable environment position shifts, such as in robotic ultrasound scanning applications.