ABSTRACT

Human behavior can never fully be predicted, posing a challenge for autonomous vehicles which rely on their prediction ability to ensure the safety of the passengers, other drivers, and, most vulnerable to injury, pedestrians. The purpose of this research is to explore and strive to improve upon traditional methods for pedestrian trajectory prediction. The social force model attempts to define human behavior by properties which are influenced by specific stimuli. This model accurately predicts trajectory in this way without neural networks, artificial intelligence, or deep learning. It describes forces from pedestrians upon one another and from obstacles onto pedestrians, but is otherwise very limited in its programmed forces. To improve this model, a hazard force which is representative of smoke has been implemented and actualized through the simulator. Although the social force model may not be the most efficient or effective method compared to deep learning models, this description of human behavior is crucial for trajectory prediction, and it may be very useful in conjunction with artificial intelligence.