Wed Dec. 8, 3:30, WeH 4601

	Semi-automatic Synthesis of a Fuzzy Logic Controller

			     Filippo Neri


The robot control function for a given application can be
semi-automatically learned in a two step process.  In the off-line
phase, the control function is approximated using fuzzy logic and
induction on a set of real traces of sensors data and associated
control action.  The result is an approximate fuzzy logic controller
that can be refined on-line, to improve efficiency, by means of a
reinforcement learning method.

A description of the whole synthesis process is presented on a
simulated task and a comparison with a manually written controller is
made. Finally, the initial studies for the scaling up on a real
application are discussed.