Analysis of Performance of Cognitive Agents Learning to Cross a CA Based Highway using Improved Learning Mechanism
This thesis presents a modification of a simulation model of cognitive agents learning to cross safely a cellular automaton based highway. Analysis and reflection focusing on the design of the initial simulation model, the mechanism of “learning” used in this model, and the shortcomings of the model are described. Based on the analysis of the simulation results of the initial model, a modification of the learning mechanism is proposed, implemented and performance of the modified simulation model is investigated. A large amount of simulation data is generated by the modified simulation model for various combinations of configuration parameters’ values. An exhaustive comparison analysis of simulation results produced by the modified and original models is conducted. These analysis involves the design of simulation experiments, data visualization, analysis of various statistics as well as the systematic comparison of clustering of histogram plots of these models. This comparative analysis of models performance conducted for various combinations of parameters brings a better understanding of the performance of cognitive agents’ models, in particular the agents decision-making processes, as well as their learning mechanisms. The thesis provides recommendations on how to improve further the simulation model of cognitive agents learning to cross a highway, their decision-making formula and the design of knowledge-based table.