Over the past decade, researchers have been developing new ways to model pedestrian egress especially in emergency situations. The traditional methods of modeling pedestrian egress, including ow-based modeling and cellular automata, have been shown to be poor models of human behavior at an individual level, as well as failing to capture many important group social behaviors of pedestrians. This has led to the exploration of agent-based modeling for crowd simulations including those involving pedestrian egress. Using this model, we evaluate different heuristic functions for predicting good egress routes for a variety of real building layouts. We also introduce reinforcement learning as a means to represent individualized pedestrian route knowledge. Finally, we implement a group formation technique, which allows pedestrians in a group to share route knowledge and reach a consensus in route selection. Using the group formation technique, we consider the effects such knowledge sharing and consensus mechanisms have on pedestrian egress times.