Simulating effects of signage, groups, and crowds on emergent evacuation patterns.

Running the simulation model

Agent Based Simulation Model

Planning of the simulation:During the simulation, the dynamic attribute values are updated at each process stage as described belowThe Perception Module updates four attributes such as:• Emergency cues, such as smoke and alarm, that are visible or audible to the agent.• Visible floor objects, such as doors and signs, that are visible to the agent.• Visible group members that are visible to the agent.• Neighboring agents that are visible to and are located within a certain radius from the agentThe Interpretation Module maps the current knowledge of the agent into a set of internal thresholds that describe the urge and well-being of the agent.The decision-making module invokes the decision tree modeling the behavior assigned to the agent. Given the agent’s characteristics and the invoked decision tree, it looks up the agent’s behavior and determines the long- term navigation goal, such as the familiar exit of the agent or the location of the group leader, and the intermediate navigation point given the agent’s knowledge and location.The Locomotion Module calculates the agent’s movement toward the navigation target and returns the updated spatial position of the agents, which areCartesian coordinates (x, y, and z) in the continuous space.The Memory Module registers the decision made during the simulation cycle and updates the spatial knowledge. The spatial knowledge is an array storing the navigation points that the agents have visited. The agents remembered the traveled navigation points and can later refer to the spatial knowledge to avoid backtracking.

Semantic representation of building safety features, visibility graph  (for the specific technical characteristics and the architecture of the model please see the article)Defined rules: Rule #1 An agent can detect the navigational points that are within the line of sight at each simulation step. Rule #2 An agent chooses intermediate navigation points based on its navigation destinations and its knowledge of the building.Rule #3 An agent ‘‘memorizes’’ the traveled space to avoid backtracking.At the individual level, an agent has a physical profile, a level of familiarity with the building, and prior known exits of at least one that the agent enters. The physical profile includes attributes such as age, gender, body size, travel speed, and personal space.• At the group level, the attributes defined for social groups include a group leader (if any), the group intimacy level (e.g., high intimacy for a family group), the group-seeking property (describing agents’ willingness to search for missing members), and the group influence (describing the influence of a member to the others in the same group). The agents belonging to the same group share the same group attributes.• At the crowd level, an agent’s social position is defined by the social order that reflects the likelihood of the agent to exhibit deference behavior. The lower the social order, the higher the chance for theagent to defer decision to other agents when negotiating the next move. A special agent, such as authority figures and a safety personnel, may have assigned roles and is responsible for executing actions, such as sharing information and giving instructions.

This paper describes an agent-based egress simulation tool, SAFEgress, which incorporates important human and social behaviors observed by researchers in safety and disaster management.

Simulation results from the demonstration indicate that occupant’s exit preferences, visual perception of the signage system, herding behavior, and social behavior among groups can lead to very different reactions to cues.The first test studies the effect of additional exit signs on evacuation performance. The total evacuation time is 165 s (averaged over 10 simulation runs). As highlighted in the figure, in this initial exit sign arrangement, agents take detours and explore the floor before find their way to exit. With additional exit signs posted, the agents travel with more direct routes and the evacuation time takes 119 s (a decrease of 28 % in time compared to that of initial layout of fewer exit signs).The second test illustrates how changing the exit orientation can help direct crowd flow. With the sign arrangement in the first test case, agents tend to exit through the main entrance and cause the congestions at the main entrance. With the proper exit orientation, more agents perceived the exit sign and its direction and evacuated through the near exit. Consequently, the evacuation time is 89 s, a further improvement of 25 %. During the simulation, the agents query the spatial model with the known exits and retrieve the shortest paths to the known exits. At the decision making stage, the agents choose to move to the visible navigation points along the shortest paths to get to their known exits.During evacuation, members belonging to a group, such as families and close friends, concerned the safety of their group members and often seek out and evacuate with the entire group even when evacuation is urgent. Often, as opposed to moving toward familiar exits, people may follow social cues and choose the exits preferred by the crowd as they observe others’ actions. We model the ‘‘following the crowd’’ behavior as follows: during the simulation, the herding agent (who is seeking to follow other agents) perceives the space and detects visible floor objects. At the decision-making stage, the herding agent assesses, for each visible floor object, the number of neighbors who are traveling toward the floor object. The herding agent chooses the visible floor object with the highest number of neighboring agents traveling toward because the agent considers the movement of its neighbors as a social cue to explore potential areas for exits. If there are no visible floor objects that other agents move to, the agent then will adopt other navigation strategies, such as referring to their known exits or following the visual cues.

With the help of SAFEgress agent based simulation model, to study the impacts of different exit signage systems within the constraints of given building layout. SAFEgress can be used to study the effects of human and social behaviors on collective crowd movement patterns.