Modelling behavior in agent based simulations of crowds

• Main Author: Vaisagh Viswanathan
• Other Authors: Bo An
• Format: Theses and Dissertations
• Language: English
• Published: 2015
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Simulation and modeling; DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; DRNTU::Social sciences::Psychology::Experimental psychology
• Online Access: https://hdl.handle.net/10356/62668
• Institution: Nanyang Technological University

Crowd simulation is gaining an increasing amount of importance in recent years for a variety of purposes ranging from movies and games to safe design of buidlings, planning for major events like the Olympics and for preparing for emergencies. Over the last few decades, crowd simulation models have come a long way from the simpler network based approaches to simulations using agent based modelling (ABM) which have a much greater amount of detail. The bottom up approach of ABM allows modelers to consider the complexity of human behavior in much more detail than was traditionally feasible. It is possible to consider thousands of individuals with their individual behavior characteristics occupying and interacting in complex indoor environments like shopping malls. However, existing computational models fail to take into consideration the large body of work on human behavior during emergencies and make unsubstantiated assumptions like perfect knowledge of the layout and immediate evacuation on hearing a fire alarm. In this thesis, the existing work on human behavior during emergency egress is studied and perception, event identification, spatial knowledge acquisition and navigation are identified as the four key building blocks of a behavior model for agent based crowd simulation. Following this, key shortcomings in existing models of each of these are identified and addressed. Existing simulations model perception as a simple process of visual sensing of a fixed circular range. A more realistic perception model considering the limitations of human information processing capacity is proposed and its usefulness is demonstrated through its effect on existing motion planning systems. Pre-evacuation behavior is generally either simplified or ignored despite studies showing its importance. A model for simulating pre-evacuation behavior is introduced and its importance in developing better strategies for egress management is demonstrated. Due to limited understanding of the impact of partial spatial knowledge on indoor wayfinding, existing models assume complete knowledge. A game-based methodology was used to reveal patterns in indoor wayfinding like decision points and the impact of short-term memory which enables more accurate modeling of evacuation. The effect that movement model choice has on egress dynamics has never been studied systematically. The final contribution is a method for quantitatively comparing motion planning systems and a comparison of three popular motion planning systems that revealed the importance of zoned evacuation time as an important metric for validation.