The Game Theory of Human-Robot Interaction

Robots can offer a solution to critical societal issues such as manpower shortage in hospitals, at home and in industries due to the aging population. Instead of replacing human workers (like conventional industrial robots), new robots should co-exist and collaborate with humans. Applications of human-robot interaction can be easily found in daily lives: robot-assisted rehabilitation, where the robot helps human patients complete a certain movement and regain various functions; collaborative object manipulation, where the robot carries an object together with its human partner and shares the object load; and semi-autonomous driving, where the vehicle's controller (robot) shares the control of the vehicle with the human driver and provides assistance to them in tasks such as line following and obstacle avoidance. The interaction behaviours in these applications range from collaboration, co-operation to competition:
Application 1 (co-operation and competition): in robot-assisted rehabilitation, a robot should provide assistance to the patient when they could not complete a task by themselves; it should reduce its assistance to and even challenge the patient to promote their learning according to their recovery progress.
Application 2 (collaboration): in collaborative object manipulation, a robot and its human partner have the same target position to reach but they may have different motion plans due to their individual local sensing of the environment, so the robot should consider the human partner's motion intention when planning its own motion and possibly the human also adapts to the robot's behaviour.
Application 3 (co-operation and collaboration): in semi-autonomous driving, a robot should be able to complete a well-defined task, e.g. track a lane in normal conditions and when necessary allow the human driver to take correcting actions by shared control of steering, e.g. changing to a new road.
This project will develop a unified framework to analyse these different interaction behaviours, and more importantly, will design a robot controller to achieve natural and efficient human-robot interaction. Differential game theory, which has been proved to be powerful in modelling multi-agent systems, is a suitable choice to categorize interaction between a human and a robot. However, how it can be used to develop a robot controller that efficiently responds to its human partner needs to be investigated. Two fundamental problems will be addressed: how to continuously identify the human partner's motion planning through haptic information and how to update the robot's control strategy to ensure a desired interaction. In this project, identification techniques will be employed to estimate the partner's motion planning and control theory will be used to develop a stable and optimal robot controller. A targeted benchmark system of robot-assisted physical training will be developed to test and illustrate the power of the proposed approach in improving the training system and predicting human behaviours.
We envision that the game theory robotic controller will enable human users to interact with a robot as intuitively and efficiently as with a human, since the robot will adapt its behaviour to the human partner according to the context of the task. This project promises breakthroughs in human-robot interaction.

A natural and efficient human-robot interaction framework has significant impacts in a broad range of domains, considering its various important applications and effectiveness. The following is a list of domains on which this project has a direct impact. In the short-term we are focussing on healthcare while maximising communications and visibility to enhance our contribution and influence across different domains.
1. Health care
The aging population challenges healthcare systems all over the world, of which one result is shortage of health service workforce. Introducing robot therapists to healthcare systems will help address this critical issue at large. The UK's House of Commons Science and Technology Committee October 2016 report said '...we are also beginning to see them [robots] working in conjunction with humans. The results to date have been encouraging, particularly in healthcare.' One ultimate goal of this project is to develop a compact robot-assisted physical training system to be used at home, thus encouraging patients to remain at their homes instead of going to hospitals. This becomes possible only if a robot can provide effective training to the patients, which is a specific objective of this project. To this extent we are collaborating with two companies GripAble and Articares who will be able to share insights on physical training and help test the interactive control algorithms.
2. Automation in small and medium-sized enterprises
Small and medium-sized enterprises (SMEs) represent 99% of all businesses in the EU. In SMEs, there are many tasks that are too complicated to be automated while too heavy-loaded to be done by human workers, e.g. small workpiece assembly and welding. The human-robot interaction framework to be developed in this project allows ad-hoc semi-automated high-mix low-volume production. The combination of complementary capabilities of human and robot increases working efficiency and production quality.
3. Extreme environment industries
The UK government will invest £93 million to develop new technologies and systems (including robotics and artificial intelligence systems) that can be deployed in extreme environments, for industries such as nuclear energy, offshore energy, deep mining and space. While fully autonomous robot systems cannot fulfil real-world requirements in many of these applications, tele-operated robots allow human users to be in the control loop and take correcting actions when necessary. The proposed framework in this project fits perfectly with the requirements of an intuitive and efficient tele-operation system.
4. Robotics skills
The UK's House of Commons Science and Technology Committee October 2016 report pointed out the importance of developing skills in robotics and AI to ensure the UK's leadership in these areas. Through this project, the PDRA and PhD student will gain skills and experience in system analysis, programming, experiment design, project management and presentation, which play central roles in enhancing their career development. As a new BSc curriculum on robotics has been developed by the PI and his colleagues at Sussex, this project will contribute to the UK's need for engineering talent, especially robotics engineers.
5. General public
Since the first day of robotics development, the general public has had a concern that one day humans' jobs will be taken by robots. This project aims to convey an important message that a robot can be designed to assist and collaborate with humans instead of replacing them. It ensures a fact that humans and robots have complementary capabilities which should be combined in a best way to maximize their effect.