Explainable AI for Multimodal Disambiguation in Triaging with Care Robotics

Summary

The growing integration of artificial intelligence (AI) and robotics in healthcare presents a unique opportunity to improve the quality and accessibility of care. However, in sensitive environments like healthcare where patient needs can be complex, it is essential for AI systems to be responsive, trustworthy, and transparent. Explainable AI (XAI) is promising in these settings, especially for multimodal, context-rich environments where robots must interpret and respond to varied data inputs, including visual, auditory, and contextual cues. This project is positioned at the intersection of AI and healthcare, addressing critical needs in care robotics by investigating interactive XAI methods that advance intuitive, dependable human-robot interactions. It aims to empower robots to assess and respond effectively to diverse user needs in sensitive care settings through triaging, with specific focus on disambiguation and explainability across diverse data sources.

Primarily, the project will develop and assess XAI methods to improve disambiguation in multimodal care environments, where complex, context-rich data is common, particularly for triaging tasks. For example, if a patient attempts to mobilise and becomes unsteady, the robot could prioritise stabilisation assistance over other tasks. Such mobility triaging will help prevent falls and reduce injuries. To achieve this in a trustworthy and transparent manner, the robot should be capable of disambiguating real-time sensory data – such as differentiating between intentional movements and accidental shifts – and providing clear explanations of its prioritisation decisions by integrating and interpreting multiple sensory modalities. This information must then be communicated to the patient in a personalised and understandable manner. By delivering these responses via conversation, patients can directly interact with robots to request clarifications and even contest their decisions. Subsequently, this data could be stored to inform future care delivery.

The methods developed in this project will be implemented on embodied robotic systems in simulated and/or real-world care settings. Using user-centred experimental designs, triaging scenarios will evaluate the robots’ effectiveness in handling multimodal disambiguation and responsiveness. Both quantitative metrics and qualitative feedback will assess improvements in interaction transparency, user satisfaction, and patient trust and comfort in care support.

The right candidate for this research should have a strong foundation in AI, particularly in XAI and multimodal data processing. Skills in machine learning, natural language processing, and computer vision are essential for addressing multimodal disambiguation challenges, while knowledge in robotic simulation environments, especially embodied and interactive systems, is highly valuable. Proficiency in programming languages like Python and frameworks such as TensorFlow or PyTorch is necessary, along with knowledge of conducting health-related user studies. Given the project’s interdisciplinary nature, strong analytical, problem-solving, and communication skills are essential for effective research dissemination and collaboration. The candidate should also have a keen awareness of responsible AI principles, as the project emphasises the ethical deployment of socially responsive and understandable AI systems in healthcare.

Ongoing research links

This project draws on our multidisciplinary research in computing and health leveraging insights from the EU SelfBack project and the XAI EPSRC project. It also fits well with the ongoing Emergence feasibility project on Robotics4Fraility with Nottingham University.

Supervisors

Discuss this further with a potential supervisor for this research degree:

Research Themes

Find other Research Degrees in the same theme: