Many applications can benefit from animated virtual crowds. These applications include site planning, education, entertainment, training, and human factors analysis for building evacuation, or other scenarios where masses of people gather such as sporting events, transportation centers, and concerts.Animating virtual crowds is often accomplished by local rules, 1 forces, 2 or flows.3 One of our objectives in crowd animation is to realistically simulate how human communication affects the behavior of individual agents. We have developed Multi-Agent Communication for Evacuation Simulation (Maces) to combine local motion driven by Helbing's model 2 with high-level wayfinding using interagent communication and varied agent roles. Together, these factors automatically augment an agent's mental map of the environment to produce empirically better building evacuation performance and realistic crowd movements.Crowd evacuation from large and complex building spaces is usually hindered by people not knowing its detailed internal connectivity. In such circumstances, occupants might not be aware of the existence of suitable circulation paths or, in case of emergencies, the most appropriate escape paths. Psychology studies show that building occupants usually decide to use familiar exits, such as where they entered the building. Emergency exits or exits not normally used for circulation are often ignored. If a fire occurs, blocking some of those known paths, and smoke further obscures vision, the problem might be fatally aggravated.In general, building evacuation due to imminent danger is accompanied by considerable physical and psychological stress. Since rising stress levels diminish full sensory functioning, there is a general reduction of awareness and increase in disorientation.Decision skills in emergency situations are influenced by several factors such as environmental complexity, dynamically changing situations, and time pressure. If people have not been properly trained, they are likely to feel stressed and might be incapable of making good decisions. On the other hand, individuals such as firefighters are trained to make decisions in a dynamically changing environment based on perception, communication, and knowledge. For untrained individuals, too much or too little information coming at one time (several people in the same room making different decisions and shouting different information about blocked rooms) can also promote indecision.Many different methods exist for simulating the local motion of individuals in a crowd such as cellular automata, social forces, and rules. These models simulate people moving within a familiar environment trying to reach their destination while avoiding collisions with walls, obstacles, and other individuals. None of the previous work in crowd simulation deals with unknown environments where agents must explore the building and communicate with each other to learn useful features and find their way toward an exit as real people would do.The main novelty of our approach to crowd ...