Modeling environmental impacts on cognitive performance for artificially intelligent entities

Abstract

The Marine Corps utilizes virtual simulations as a training tool for ground combat operations. Currently, the artificial intelligence of simulation entities does not exhibit appropriate performance degradation due to environmental conditions, such as heat and humidity. To address these gaps, this thesis reviews existing approaches to modeling the influence of environmental factors, specifically heat and humidity, on human performance in vigilance and attention tasks. We present a novel agent behavior model that incorporates a modified A* search pathfinding algorithm based on empirical evidence of human information processing under the specified environmental conditions. Next, an implementation of the agent behavior model is presented in a military-relevant virtual game environment. We then outline a quantitative approach to test the agent behavior model within the virtual environment. Results show that our human information processing–based agent behavior model demonstrates plausible agent performance degradation in hot, humid temperature environments when compared to paths around the danger area in mild temperature environments. The results of this research provide an approach for implementing an agent behavior model that accounts for environmental impacts on cognitive performance. We recommend future work to validate the model in a human subjects experiment to facilitate improving the realism of simulation training.