Research Line 3: Probabilistic planning
- Coordinator: Leliane Nunes Barros
- Group members: Fabio G. Cozman, Anna H. Reali Costa, Hector Geffner, Silvio do Lago Pereira, Jeronimo C. Pellegrini, Karina Valdivia Delgado, Felipe Werndl Trevizan, Fabio Natanael Kepler, Ricardo Shirota Filho e Nicolau L. Werneck.
The area of probabilistic planning in Artificial Intelligence (AI) and the decision-making under uncertainty of Operational Research (OP), while solving similar problems, apply different methods. The area of planning in AI assumes that problems are specified by a language of high level, based on probabilistic logic, used to describe the world's states and actions probabilistic (probabilistic transitions states), and efficient techniques to solve these problems make the assumption that the agent's goals are goals reachability of states, which defines a stochastic minimum path problem, and that they are reachable from any state of the world.
On the other hand, OP aims at building mathematical models, such as a Markov Decision Process (MDP) and obtaining efficient algorithms to optimize an objective function. However, both fail when trying to solve practical problems involving sequential decision making in general, due to the following: a large number of states and actions; imprecise probabilities on the state transitions; and more general goals of optimization. The goal of this line of research is to obtain compact forms of representation spaces of states involving imprecise probabilities for planning problems with more general goals.