Evolutionary multiobjective optimization for memory-encoding controllers in the artificial ant problem

Many agent problems need efficient controllers that the agent takes to handle the environmental information. If the sensor information about the environment is limited, dynamic processing of internal memory is required. An agent solves the artificial ant problem with internal memory, where an agent is supposed to collect all the food pellets on the trails. In this paper, we provide an evolutionary multiobjective optimization approach to quantify the amount of memory needed for desirable behavior performance for the agent problem. For the approach, we use finite state controllers to encode internal memory. The approach uses two objectives, number of internal states and behavior performance. The goal is to maximize the behavior performance of the agent with each level of internal states. The suggested method with elitism strategy can find efficiently desirable controllers for the artificial and problem.