OPTIMISATION DE PROCESSUS DECISIONNELS POUR LA ROBOTIQUE

Abstract

On the basis of the general formalism of "Production Systems" two types of Decision Porcesses are defined: closed processes for finite state-space systems, and Open processes for infinite state space systems. The thesis considers such processes as particular algorithms, and deals with their modelization, analysis and optimization of their complexity for various criteria and cost models, taking into account the complexity of the optimization task itself. The task is characterized as NP-hard in the strong sense, and carried ouot by some original approximation schemes, developped as a generation of admissible heuristic search procedures to near admissible one. Two Robotics decision processes are considered : synthesis of optimal classifiers for object recognition, and plan generation.