Adaptivt lärande för personifierade utbildningsrobotar
Tidsperiod: 2016-01-01 till 2019-12-31
Projektledare: Ginevra Castellano
Budget: 3 452 000 SEK
After the advent of industrial automation, we are now witnessing a second robotics revolution, but, as scientists strive to move the robots out of the factories and better integrate them into our daily lives to support and instruct humans to complete tasks, robot adaptation and personalisation abilities remain limited. A key requirement for an instructional robot is to adapt to its users by providing them with ad hoc support and instructions to maximise their performance during a given task and create an engaging user experience. To date most adaptive instructional robots rely on pre-scripted rules, but do not adaptively learn from the effects of such strategies, for example if they have been effective for a specific user in a specific context. This project will develop novel, ground-breaking adaptive learning algorithms for effective human-robot interaction in instructional settings that rely on robust real-time modelling and adaptation to human users and task performance, delivering highly personalised robots. To ground the research in a concrete instructional context the project will develop and evaluate new reinforcement learning approaches in the area of robot-supported learning, with a specific focus on social robots acting as educational agents. If successful, the project will open up new frontiers and opportunities, making a first step towards the development of robots that achieve long-term robot autonomy, prerequisite for a successful integration in our daily lives.