Trust-based variable impedance control of human–robot cooperative manipulation

Human–robot collaboration (HRC) systems integrate the strengths of humans and robots to improve joint system performance. In particular, human–robot cooperative manipulation (co-manipulation), a prominent area within the field of HRC, in which humans and robots manipulate the same object, has garnered significant attention. Trust in HRC is crucial in determining the level of human acceptance of robots and, hence, robot utilization. This paper develops a probabilistic dynamic Bayesian network (DBN)-based trust model and trust-aware variable impedance control for human–robot co-manipulation. Due to the continuous nature of trust evolution and the limitation of the classic parameter learning approach, a continuous and normalized version of Baum–Welch (BW) algorithm is developed to learn the trust model. Trust is vital in any HRC task, such as human–robot co-manipulation. To achieve successful co-manipulation, it is necessary to take into account human–robot trust and further facilitate seamless HRC. Therefore, we propose a comprehensive framework with trust-aware variable impedance control of robot behavior for human–robot co-manipulation, based on human trust in robot and force-based human intention estimation, with trust-based obstacle avoidance and trust-based robot-level task hierarchy. An extensive case study on human–robot co-transportation is conducted, encompassing five distinct robot behaviors, namely, compliant, autonomous, switching control-based, variable impedance control based only on human applied force, and our proposed trust-based variable impedance control behaviors. A rigorous statistical analysis is performed to test their statistical significance and our trust-aware variable impedance control strategy demonstrated its superiority in terms of efficiency, agreement, safety, pHRI, and sHRI factors compared to other benchmark behaviors. The statistical findings also indicate that our proposed trust model has the ability to reflect human subjective surveyed trust qualitatively.