By integrating counterfactual reasoning with advanced natural language processing (NLP) techniques, we aim to improve how care robots explain their decisions to people whom they are supporting. Counterfactual reasoning allows robots to describe how altering certain conditions could lead to different outcomes, making their actions and limitations clearer and helping to build trust between robots and users.

Our approach involves creating an NLP robot-care simulation framework, using co-design activities, that incorporates NLP to produce counterfactual explanations from real-life interactions between robots and older adults. This framework will combine methods in Natural Language Understanding and Generation to process human natural inputs and articulate explanations. To ensure effectiveness in actual care settings, we will test and refine our prototype within the simulated environment.

This research addresses a vital gap by equipping care robots with the capability to transparently share their reasoning processes. Ultimately, our goal is to enhance care management for those experiencing frailty, making robotic aides more accessible and supportive, while also enhancing user trust and acceptance. 

Findings from this seed funding initiative will inform the identification of key theoretical and practical challenges, laying the groundwork for a joint UKRI proposal involving RGU, Orkney, and Nottingham, aimed at enhancing technology to better meet the needs of older adults with embodied robotic aides.

Our methodology is organised into 4 work packages based on 4 key methods described below. 

Co-design Activities (Initial 4 Months): 

Four structured co-design activities with older adults are planned with the following aims.

1. Focus Group Discussions: Aim here is to gather initial perceptions about embodied robots in care settings, identifying assistance areas, expectations, and concerns. We will also inform older adults about the type of technology that is being considered. 

2. Care Task Identification: Aim is to prioritise specific care tasks through simulations and discussions, focusing on those most relevant to the target user group. Through a discussion, we will arrive at a consensus as to what is actionable within this project.

3. Role-playing Scenarios: Aim is to simulate selected care task interactions, collecting data to inform dialogue model development. This will be achieved through several simulated scenarios to help manage role-playing activities.

4. Enhanced Simulator Feedback: Aim is to obtain feedback on the prototype focusing on usability, acceptability, and overall satisfaction with the simulated scenarios.