Fuzzy anticipatory learning classifier system for mobile robot navigation

Designing autonomous intelligent control systems for real-world problems is a daunting task. The complex input-output relationships resulting from the interaction between a process and its environment are often not readily solvable by traditional mathematical methods. A growing amount of research is being performed in designing control systems which develop their own solution by utilizing methods borrowed from nature. This thesis presents work performed in the aforementioned field, specifically in developing an extension to the Anticipatory Learning Classifier System (ALCS) to facilitate the transparent use of real-valued inputs as well as outputs in order to make the system more applicable to real-world problems. This has been accomplished through the application of concepts borrowed from Fuzzy Logic to implement a variation of an evolvable Fuzzy Controller within the ALCS paradigm. As such, the Fuzzy Anticipatory Learning Classifier System (or FALCS) allows the user to evolve an adaptive control system capable of latent learning as well as utilizing the best known course of action in the absence of previous knowledge. The FALCS-based controller was tested to be successful in generating a rule-base that kept a simulated agent "alive" in a virtual environment. Furthermore, a FALCS-based controller was successfully implemented to allow a simulated robot to navigate a previously unknown environment, as well as seeking a goal location while avoiding obstacles at the same time.