About the Project

Project description

Robot Autonomy will require the systems or robots to set up their own agenda (in line with the tasks they are meant to do), defining the next goals to achieve and discarding those who can’t be completed. However, this may create misunderstandings with the users interacting with the system, who may expect something different from the robot, or may be confused if the robot behaved in a non-legible manner.

Therefore, it is important that these autonomous systems are able to explain why they achieved one task and not another, or why some new (unexpected) task was achieved that was not scheduled. Other sources of misunderstandings may come from action failures and replanning, where the robot finds a new plan to complete an on-going task. In this case, the new plan may be different to the original one, thus changing the behavior that the robot was performing.

This project will explore how to generate goal-based explanations for planning-based robots in assistive/service scenarios, extracted from goal-reasoning techniques. It will also look at plan repair to enforce cohesion after a replanning to ideally increase the trust and understanding of the users about the system. Those explanations should also contemplate unforeseen circumstances, therefore explaining things based on “excuses” that the robot may give to the user. Finally, we will investigate how to obtain and provide those explanations at execution time, so explaining on-the-go. The methods developed shall be integrated into a robotic system, in an assistive/service robot scenario.

The methodology will be based on AI Planning techniques, where task domains are represented in PDDL. The project will first delve in the literature of goal generation and reasoning for robots, and explore the explainability opportunities in such methods. Afterwards, it will explore extensions to such methods to generate better goals, ideally being self-explanatory. The project will evaluate the different methods with both objective metric-based assessments and user study that will focus on trust enhancement and understanding of robot’s motives.

Explanations are crucial in Human Autonomy Teaming, as they have been proved to enhance trust. This work is highly novel, both in terms of goal-based explanations and generation of goals for planning-based robots.

Programme details

The successful candidate will be registered on the Computer Science Research MPhil/PhD in the Department of Informatics at King’s College London. They will be part of the King’s Prize Doctoral Programme in Safe, Trusted and Responsible AI.