This paper presents machine learning concepts to successfully implement an intelligent agent solution to the problem of autonomous control of Personal Rapid Transit (PRT) vehicles. The PRTs can operate independently and together as a whole system. Thus, they can be considered as a system of systems. By developing a set of rules for the vehicles to follow, the various parts of a Bayesian Network (BN), its nodes, parameters, structure and utility function are determined. The variables in the network are generated from the sensor data of the PRT vehicles. Using the overall system goals, a utility function can be devised that will select the best action for an agent to take. Simulation results show that this methodology is useful for "rapid prototyping" a decision theoretic agent. The gap between the intelligent agent and rule based agent performance for the collision metrics suggests that more research is needed to develop a systematic method of creating utility functions to meet system goals.