This is the ARPIA architecture:
In short, we need the acquisition of information from sensors, which will be pre-processed in order to extract the core data. This data will be handled by the learning techniques to derive behavioural models. These models can be built from scratch (completely learned from the input observations) or existing models can be improved with new observations. These models will be used later by the behavior recognition techniques to interpret new observations and to predict the agent’s behavior. With such input on the identified activity/plan/goal (behavior), recommendations about the next action to perform or how to reach a specific goal will be given.
Four proof-of-concept applications related to traffic, tourism, robotics and domotic systems will be studied in the project.