The Artificial Intelligence group is one of the research groups of the Department of Information and Communication Technologies at the Universitat Pompeu Fabra (UPF) in Barcelona, Spain, as well as a member of the Center for Autonomous Systems and Neuro-robotics (NRAS).
The Artificial Intelligence group
The group research covers a number of areas in AI including planning, reinforcement learning, and constraint satisfaction. Planning is the model-based approach to intelligent behavior, where behavior is not programmed or learned, but is derived effectively from a model of the agent goals, and the way that the actions and sensors work in the world. Reinforcement learning, in its standard form, is a model-free variant where behavior is derived from experience by maximizing expected rewards. Finally, constraint satisfaction deals with the problem of finding assignments of values to variables so that a given set of constraints is satisfied. These three areas are essentially related to models of actions, inference, and learning, and are also related to each other.
One of the key challenges in each of the previous areas is computational: how to scale up planning, learning, and inference methods in order to deal with problems involving a large number of variables. We approach this challenge in two ways: theoretically, by understanding what makes these models easy or hard, and experimentally, by developing methods that can automatically exploit the structure of the problems. The tools for carrying out this research thus combine logic, probability theory and complexity theory on the one hand, and heuristics, algorithms, programming, and experimentation on the other. The group is known for a number of general, domain-independent solvers that we have developed and tested over the years, and for the ideas that underlie them.
The group is currently engaged in a number of multidisciplinary research projects funded both by the European Union and the Spanish Ministry for Science and Innovation, and ranging in subject from the application of planning to agent-based social simulation environments (Simulpast) to the development of goal-recognition probabilistic methods as a tool for assisting pedestrians navigate urban environments (Spacebook).
DTIC, Universitat Pompeu Fabra
C/ Roc Boronat 138, 2nd floor.
08018 Barcelona, Spain.
info-ai <at> upf <dot> edu