Analysis and Simulation of Computacional Mechanisms for Action Selection

Action selection is one of the basic problems studied in Artificial Intelligent Systems. Many of the solutions involve animal inspiration. An animal in its habitat takes several decisions all the time, using as a basis their sensory perceptions, motivations and instincts. Artificial mechanisms of action selection allow artificial creatures, such as robots or characters in computer games, to make these decisions autonomously. This project aims at analyzing and studying Computational Mechanisms for Action Selection used in cognitive architectures. We focused mainly on the algorithms described in the doctoral thesis of Toby Tyrrell (199)1. The work of Tyrrell analyzes four different action selection mechanisms: the mechanisms based on drives, the Maes algorithms, Lorenz and Rosenblatt & Payton. In this work, we studied the strategies mentioned by Tyrrell (199)1 in order to reproduce two of these mechanisms, the mechanisms based on drives and Maes algorithm, and thus compare the results with those obtained by him. The algorithms were encoded using the Java language and made available as a part of the CST Cognitive Systems Toolkit, a software toolkit being developed by Prof. Gudwin's research team at FEEC/UNICAMP.