I. Introduction
An emergency happening in crowds, such as a fire or an explosion, may lead to a large number of injuries and deaths. Recent studies and investigations performed after crowd disasters suggest that well-trained agents, such as fire wardens and firemen, can significantly improve the survival rate in such crowd disasters [1], [2]. However, although the agents can guide evacuees to the closest exit, they may not be aware of the congestion at that exit, which may delay the evacuation. To avoid heavy congestions at exits, evacuation flows to all exits should be carefully organized. With recent rapid development on sensors and network technologies, it is possible to install cameras or other surveillance sensors widely at exits in buildings, if not existing already. With such surveillance capabilities, we believe intelligent emergency evacuation robots can offer evacuees better guidance with the updated environment information, compared to traditional guidance from stationary signs or agents.