Consortium: University of Lincoln and University of Nottingham

Duration: 2025

This project is funded by the UK-RAS Network Low TRL Research Activities scheme to address the fundamental challenge of how robots can learn effectively from unstructured data generated through interactions with non-expert users in real-world environments.

As robots become more common in everyday human environments, learning effectively from unstructured interactions with non-expert users is a critical challenge. While robots can be pre-programmed with specific behaviours, real-world deployments inevitably encounter situations where these capabilities are insufficient. Given that such failures in dynamic human environments are unavoidable, novel theoretical frameworks are needed to enable robot learning and adaptation from in-situ interactions.

PLUS-HRI will develop a novel theoretical framework for uncertainty quantification, intention inference, and transfer learning in these unconstrained settings. Building upon previous work with a mobile robot deployed in a public space (“Lindsey: the tour guide robot” project ), the project will use autonomous navigation as a case study. The goal is to develop new methods for learning from suboptimal demonstrations and in-situ data from human-robot interactions. The project will leverage existing data and evaluate its theoretical framework through a new real-world Human-Robot Interaction (HRI) deployment.

Solving these challenges has broad implications for increasing autonomy in real human environments, increasing interpretability and usability, and democratising access to robot control. These span disciplines beyond Robotics and AI; as such, planned follow-on activities will include multi-disciplinary perspectives to explore the theme more widely

Contact: Francesco Del Duchetto