FREEpHRI: Flexible, Robust and Efficient physical Human-robot Interaction with iterative learning and self-triggered role adaption

Globally, the number of robots in use in 2020 is over 2.25 million, which will multiply even faster in the next 10-years, reaching 20 million by 2030. The fast adoption of robots has contributed 10% of worldwide total GDP growth in the last five years. The technological advancements are bringing robots to humans' daily lives, and they are no longer working in an isolated environment, but sharing the same workspace and physically interacting with humans, e.g., rehabilitation robots, tele-operation robots and collaborative robots.

The physical coupling between humans and robots, often termed as physical human-robot interaction (pHRI), facilitates new human performance capabilities and creates opportunities to explore the task sharing and the control between humans and robots. To maximise the benefit of human-robot system during joint tasks, the robot needs to understand what the human is trying to do, and intelligently adjusts its behaviour according to the performance of the human and the requirements of tasks, which requires novel tools to model the human behaviours and innovate strategies to modulate the control of the robot.

The ambition of this fellowship is to enable robots to real-time estimate human behaviours, intelligently detect the changes of human behaviours, automatically adjust the relationship between the human and the robot (from collaborative to competitive), and provide natural interaction behaviours even when the robot dynamics are partly unknown. I will pursue this goal by: 1) developing a flexible and robust pHRI control strategy.

The control strategy uses a two-player differential game to model human-robot interaction behaviours, and learning techniques to compensate the effects of unknown dynamics and external disturbances. A cost function implying motor capability will be assigned to the human partner, and the robot will adjust its role (collaborator or competitor) according to the real-time estimation of the human cost function. 2) introducing an efficient self-triggered role adaption mechanism.

The triggering mechanism uses the performance of the human-robot system and the estimated human behaviour to detect the role changes of the human, and triggers the robot to change its role when necessary; 3) evaluating the reliability and functionality of the proposed techniques through an exemplar application in physical robot-assisted rehabilitation. The proposed techniques will be used to achieve typical training strategies (e.g., passive, assist-as-needed, challenge-based) initially in laboratory settings, and then in the Leeds Teaching Hospital rehabilitation service.

This fellowship targets at two fundamental issues in pRHI: (1) how to efficiently update the robot's control strategy to ensure desired interactions; and (2) how to deal with uncertainties in the human-robot system. The technologies developed in this fellowship will provide a general framework for designing an interactive robot control system, which has a large group of applications in both healthcare and manufacturing.

The fellowship objectives and milestones will be delivered collaboratively with partners from the University of Leeds, the University of the West of England Bristol, the University of Manchester, Leeds Teaching Hospitals NHS Trust, Devices for Dignity, YIRUIDE Medical and DIH/Hocoma.