Students

Abstract:
Although more and more AI agents can explain their behavior, they do so without considering the specific individual they interact with. To make explanations and the subsequent interactions more natural and effective, we propose to make agents aware of their user's mental and affective state by attuning the explanations they give to physiological and social cues exhibited by the user.

Abstract:
This project will investigate service robot adaptation across two time scales: short term scale, where the goal is to adapt to a single user, and a long-term adaptation framework, where individual user adaptation is reduced and optimised through discovering patterns and commonalities between different users. The particular focus of this project will be assistive robotics, where tasks like feeding and shoe dressing will be investigated.
The proposed output of this project is a learning framework and a behaviour adaptation algorithm that builds an interaction model with the user by inferring beliefs about the user’s abilities and physical properties from ad-hoc collaboration (table-top assistance, putting a shoe on). It then uses this learned model to update its global beliefs about any future interaction based on key insights from the single interaction. This should allow the framework to operate at very long timescales without need for recalibration or wiping of memory or experience. This work will also be underpinned by physical experiments, where a service robot will perform in a variety of scenarios that prove the framework design.