The speaker discusses using reinforcement learning models to simulate human-like pedestrian movement, and how their simulator differs from other simulators by solving the computer graphics problem instead of the computer vision problem.