We first introduce a cognitive architecture: CLARION (Connectionist Learning Adaptive Rule Induction On-liNe) and its comprising subsystems. In this thesis, we focus on the ACS (Action-Centered Subsystem) which plays the controller role in the overall CLARION architecture. For this focus, we first study the theoretical aspects of the ACS including its two level components: ARS (Action Rule Store), IDN (Implicit Decision Network) and respective representations and learning processes; how they are combined and how each network are coordinated; the goal structure, working memory and their roles in CLARION. Then, we discuss the overall implementation structure of the CLARION architecture briefly and the ACS implementation in great details. Following this, some experimentation (Process Control Task) only involving ACS is presented and discussed. At the end, comparisons with other cognitive models (ACT-R and SOAR) are discussed in details and the future work is pointed. |