P1. Lifelong Decision Making for Social Robots in Smart Homes
Principal investigator: Fernando Fernández Rebollo
► Coordination and management of the whole project through the methodological guidelines defined in Section C2.4. In this sense, in this sub-project, the first goal will be to define the use-cases or specific scenarios that will be implemented, and that will end with the pilot execution and evaluation supported by the medical staff. Fernando Fernández will be in charge of these goals.
► Development of the high level decision making module for the CORTEX architecture, based on Automated Planning and Reinforcement Learning approaches, which will require the formalization of the use-cases to be developed during the project in standard languages like the Planning Domain Description Language (PDDL). Both researchers will be in charge of this goal.
► Study and development of Lifelong learning approaches for decision making. In this sense, in this project we propose: a) the use and development of Policy Reuse algorithms for transfer learning in Reinforcement Learning, which has proven to be a successful mechanism to acquire policies in new tasks and domain structure (Fernando Fernández); b) the lifelong learning of different types of knowledge that can be used in different points of the high level planning, from control knowledge or domain knowledge that can support the planning process, to user and goal preferences that can improve the user’s experience (Angel García)