multi-Sensor Human Activity Recognition and intention Prediction (SHARP)

When humans and robots are present within an environment (e.g., a food factory), both types of agents must adapt their behaviour to the presence of the other. Humans seem to naturally adapt over-time; whereas robots' behaviour currently tends to be more ridged. Our aim is to evaluate how humans are affected by the actions performed by a robot, thus leading to humans and robots more effectively co-existing and cooperating.

This project focuses on the first step towards this vision, that is, on recognising the humans' activities and predicting their intentions. For example, the system could recognise a human picked-up a mop and filled a bucket with water, and thus predict their intention is to clean the floor.

The project will involve four core technology-based stages: (1) setting up an emulated agri-food environment with a set of sensors (e.g., the Vicon system); (2) designing and performing experiments to evaluate the performance of activity recognition methods within that environment; (3) combining the best performing activity recognition approach with an intention prediction method and evaluating the performance within the test environment; and (4) investigating how the approach could be scaled and deployed within a more realistic setting (e.g., the industry partner's site and/or the environments available at UoL's Riseholme campus).

The PhD candidate will have the opportunity to learn how to use and setup state of the art sensors, and how to design and conduct experiments with human participants. They will get the opportunity to work with robotic platforms and will expand their knowledge of developing activity recognition and intention prediction algorithms. They will develop their oral communication skills thorough giving talks to the research group and at conferences/workshops and their written communication skills by producing publications.