A bio-inspired group evasion behavior

Group evasion behavior is the evasion pattern of numerous agents, one kind of the most frequently observed behavior patterns in nature. However, there appears to be rather little research about human crowd evasion behavior. In this paper, I introduce a new model for simulating group evasion behavior based on biological and sociological models for the purposes of simulating the crowd animation in evasion situations. In biology, fish school group evasion behavior is well studied and it has a possibility that the human crowd evasion behavior in emergency is similar to fish school group evasion patterns. I take this biologically inspired model and seek to extend it by integrating sociological factors found in human groups. For the immediate dangers, bio-inspired model could simulate human evasion movement. With sociological factors, I can simulate more complex evasion patterns with considering sociological factors. With this model, I would be able to simulate crowd evasion behavior in emergency situations. Introduction: The collective behavior of human crowds is one of the most interesting topics for numerous researchers and research areas such as sociology, virtual reality, computer graphics, robotics, psychology, politics, transportation, etc. However, it is difficult to simulate crowd behavior with a good mathematical model because many factors influence how each individual will behave and affects the overall crowd behavior. The factors that induce a variation of a person’s behavior are physical attributes, personality, social status, relations with other individuals, etc. These factors make scientific experiments very hard, and generate difficulties for validation. There are various models to explain human crowd behavior in sociology and psychology [27, 28, 29, 32], but none of them is proven because of the previously mentioned difficulty in the modeling and the validation. Moreover, we cannot experiment on human crowds to produce collective evasion patterns, so it is difficult to retrieve relevant data. Also, collective evasion patterns are usually expressed in emergency situations, so few pieces of video data for this behavior are available. 1. Previous works in crowd simulation and motion planning Crowd simulation and multi-agent motion planning have been studied by many researchers. One of the main research topics in robotics is to enable multi-robot agents to navigate autonomously and collision-free in various environments. In computer graphics, motion planning with autonomous agents could be used for digital actors, which can react and adapt to high level directives in dynamic environments. Navigating in dynamic environments [2, 3, 4, 6, 8] and efficient collision avoidance [1, 5] for numerous agents are main topics for multi-agent motion planning. Crowd simulation is very difficult not only for the multi-agent motion planning problem, but for the inherent complexity in the behavior of each human individual. For the purpose of handling very large number of agents, there are two major approaches for crowd simulation modeling; considering the crowd as particles or considering the crowd as a large group of agents. The particle approach for the crowd modeling is based on a continuum perspective for the crowd, which considers crowd as some continuous material. It considers crowd motions as per-particle energy minimization [5, 8, 14]. Because the system controls all the particles entirely, all the individual motions could be guided easily by global motion planning. Also, it is quite efficient for computation, because increasing the number of agents is just condensing the density of particles. However, because it is based on continuum perspective, the motions of individuals are similar to fluid and not quite realistic [5, 14]. The multi-agent approach for crowd modeling is much more complex than the particle approach. The complexity of the multi-agent motion planning increases exponentially with the number of agents and their degree of freedom. However, with this model, each agent can react with some intelligence, similar to the real world, so we can control each agent’s movement more precisely with more details and with autonomous navigation capabilities [2, 3, 4, 6, 8]. Within various group behaviors, the crowd evasion behavior is useful for numerous applications: animation, simulation, social science, virtual reality for various situations: evacuation, police chasing to arrest the riot or protester situation, minimizing damage in terror situation, escaped animals from zoo, controlling disorder situation, or video games. There is some good research on crowd evasion situations evacuation [9], and suicide bombing terrorist [11]. Helbing presents collective crowd behavior of evacuation, induced by panic, using a simulation based on a social force model. He considered the social force with psychology information, and simulates group evasion patterns resulting jamming in life-threatening situations. Zeeshan shows that running to an exit exposes a person to death threatening situation when the suicide bomber attacks the crowd. However, we are not aware of any works in crowd simulation about use of evasion behaviors based on biological models. 2. Bio-inspired algorithms It is a well developed approach to introduce biological algorithms in various fields: optimization, robotics, networking, social organization, etc. In particular, robotics researchers are mainly focused on an emergent behavior of a biological swarm, which is a high-level goal-driven group behavior resulting from the cooperation of simple individual patterns. Emergent group behavior patterns are easily found as insect swarm and animal herd behaviors in nature. Biological individuals only have limited sensing capability, and use simple and robust decentralized algorithms. Biological algorithms are easily applied to simple autonomous robots for these properties, so there are numerous bio-inspired research studies in robotics [17, 18, 20, 23, 24]. Because the biological algorithms or procedures have survived the evolutionary process, they are shown by natural selection to be effective, robust and efficient. Bio-inspired approaches are often used when dealing with overly complex problems, or when there exist similar problems in nature. Many researchers have used biological inspirations. Reynolds [15] simulates the flocking behavior inspired by flock of birds. Svennebring and Koenig introduce ant pheromones for their terrain covering robots. Schwager et al. bring in the ladybugs’ algorithm to solve terrain coverage problem with distributed agents [20]. Halasz et al. introduce the quorum sensing for the redistribution of swarm robots, which is inspired by ants’ house hunting algorithm [18]. Stafford et al. utilize the locust vision system in the collision detection mechanism for cars [24]. Barrows adopts the optic flow to UAV flight control, which is used by insects’ vision to avoid collision [23]. The genetic algorithm and the ant colony optimization are used in numerous fields: game theories, NP complete optimization problems, etc [19]. However, there are few papers [15] in computer animations for group behaviors which are biologically inspired. In emergency situations, human crowd reactions are similar to animal group behavior in a high level perspective, because of the urgency to react – “Individuals start pushing, and interactions among people become physical in nature” [9]. The reactions of people vary at the different levels of danger and situations [29]. However, with immediate dangers for primitive sensors, people usually have no time to think about the complex environmental information, so I believe that there should be some linkages between the crowd evasion behavior and the animal group evasion behavior. Therefore, adopting animal group evasion models should be useful to simulate the crowd evasion patterns. Animal group evasion behaviors are well studied because it is one of the most important behaviors for animal researchers. Unlike the human crowd evasion behaviors cases, the experiment for the animal group evasion behaviors easily controls variables, and generates repeatable situations. Therefore, animal researchers have developed good mathematical models for animal group evasion cases. Especially, fish group evasion patterns are one of the best developed behaviors and have some good mathematical models with relevant evidence data [21, 22]. Among the fish group evasion behavior models, Inada et al. [21] suggest a good mathematical model with the validation of simulation and proper proof. Its basic model is based on Aoki’s and Huth & Wiseel’s model. This model is based on individual behaviors, so it is proper to generate emergent behaviors. It is two-dimensional model based on velocity and angular change with decision time delay also being considered. The decisions of fish individuals are based on motions of neighborhood, with some tendency value. The basic strategy for grouping is similar to Reynold’s flocking [15]. Fish individuals move during building a group with motions of approaching, parallel orientation, and repulsion. With this model, the authors could simulate all reported evasion patterns of real fish schools, except for the ball pattern. It is a good model with reliable validation data, but has some limitations. Their model only considers free field with no obstacles, and only one predator. Also, they do not think about the factor of domains of danger [25] or the speed change for urgent evasions.