Automatically Improving Agents Behaviors in an Urban Simulation

Our goal is to create realistic urban simulations involving pedestrians, cars, pedestrians crossings and many others urban agents. These simulations help architectural designers in choosing architectural configurations. A problem related to this simulation is to create agents that have realistic behaviors and that are also efficient (a simulation may manage thousands of agents at the same time, so modeling an agent's behavior has to be rapid). Therefore, we have developed a program that automatically improves the agents behaviors given (1) some simple situations to avoid (a car that run over a pedestrian, or a pedestrian that tries to walk on another one) and (2) the rules of the simulation. The rules that describes the world and the rules describing the situations to avoid are written using predicate logic. The program that automatically writes the agents is written using metapredicates that manipulates the predicate logic rules describing the simulation. The advantage of creating the agents automatically is to have a lot of reliable, efficient and quickly designed rules.