A Cognition-Emotion-Personality Framework for Modeling Human-Like Awareness and Behavior in Emergency Evacuations

Problem Statement
Traditional agent-based evacuation simulations often rely on simplifying assumptions that limit their ability to reproduce complex, heterogeneous, and non-optimal behaviors observed in real emergencies. Common limitations include the assumption that agents possess perfect event awareness, complete knowledge of exit locations, and fully rational decision-making capabilities. These idealized scenarios fail to capture critical real-world phenomena such as pre-evacuation delays, confusion, disorientation, panic-induced injuries, and the spread of fear. Consequently, there is a need for a unified framework that integrates cognitive, emotional, social, and personality-driven mechanisms to model human behavior under uncertainty more realistically.

Methodology
The paper proposes an extended agent-based evacuation framework that builds upon and modifies the ESCAPES simulation environment. The methodology introduces a unified cognitive-behavioral model where agents possess internal states that dynamically influence their perception, decision-making, and physical interactions. The core components of the framework include:

1. Dynamic Event Awareness: Instead of discrete levels, the framework uses a continuous Event Certainty Level (ECL) in the range [0, 1]. Agents accumulate evidence through direct environmental perception (seeing the fire) or social interaction with neighbors. Evacuation is triggered only when an agent's ECL exceeds an individual-specific threshold, capturing heterogeneity in responsiveness to risk.
2. Memory-Based Exit Knowledge: Exit knowledge is modeled as a structured, dynamic process subject to acquisition, forgetting, and recall. Agents maintain a set of exit entries with attributes for knowledge status, temporary forgetting, importance weight, and last visual confirmation. Forgetting is probabilistic, influenced by the agent's fear level, time since last visual contact, and agent type (e.g., staff vs. first-time visitor).
3. Continuous Fear and Panic Modeling: Fear is represented as a continuous variable [0, 1] driven by multiple sources: fire exposure, informational uncertainty, and stagnation (lack of progress). Panic emerges as a high-intensity state when fear exceeds a critical threshold. Unlike previous models where fear primarily affects speed, this framework links fear directly to perceptual degradation, memory impairment, loss of motor control, and the emergence of physical collisions, falls, and trampling.
4. Personality-Driven Modulation (Neuroticism): The framework integrates the OCEAN personality model, specifically focusing on the Neuroticism trait. Rather than defining new behavioral rules, Neuroticism acts as a modulation factor that systematically affects fear generation, escalation rates, susceptibility to emotional contagion, and the duration/effectiveness of panic recovery.
5. Social Contagion and Interaction: Information and fear spread through local interactions. The influence of a neighbor depends on their certainty level and the source of their information (direct vs. indirect). Emotional contagion is modeled as a gradual convergence process modulated by the receiving agent's Neuroticism.

Key Contributions
The paper identifies the following primary contributions:

• Unified Framework: Integration of event awareness, exit knowledge, emotional dynamics, and personality into a coherent model linking perception, cognition, emotion, and behavior.
• Continuous Representations: Transition from discrete to continuous domains for Event Certainty and Fear levels, enabling smoother belief updates and behavioral transitions.
• Dynamic Memory Modeling: Introduction of a structured exit knowledge representation that separates acquisition from recall, explicitly modeling stress-induced temporary forgetting and heterogeneous memory robustness based on agent experience.
• Graded Social Influence: A continuous mechanism for knowledge and fear transfer where influence strength varies based on the source's certainty and information type.
• Multi-Component Fear Model: Decomposition of fear into fire-induced, uncertainty-induced, and stagnation-induced components, combined via a dominance rule to determine emotional state.
• Personality as Modulation: Implementation of Neuroticism as a modulation layer affecting fear amplification, contagion susceptibility, and recovery dynamics, allowing for heterogeneous emotional responses without discrete behavioral categories.
• Physical Consequences of Panic: Explicit modeling of how panic degrades motor control and perception, leading to emergent crowd phenomena such as falls, trampling, and injuries.

Experimental Results
The framework was evaluated through six controlled simulation experiments in a two-room building layout with 30 agents per run (30 runs per configuration). Key findings include:

• Baseline Comparison: Compared to an omniscient, rational baseline, the proposed cognitive-emotional model resulted in significantly lower evacuation efficiency, delayed evacuation decisions, and the emergence of realistic behaviors like confusion, information seeking, and leader-following.
• Spatial Familiarity: Higher spatial familiarity (Staff vs. First-Time Visitors) correlated with higher evacuation rates, lower confusion, and fewer injuries. Interestingly, high familiarity led to lower average event certainty in some cases because staff evacuated quickly, reducing social information propagation to others.
• Memory Robustness: Increasing staff experience (modeled via lower forgetting probabilities) linearly reduced the incidence of confusion and cognitive paralysis.
• Decision Thresholds: Higher evacuation thresholds (representing cautious individuals) led to delayed responses, lower escape rates, and increased fatalities.
• Fear Dynamics: Enabling fear and panic mechanisms significantly reduced evacuation efficiency and introduced physical hazards (falls, trampling, injuries) that were absent in fear-disabled simulations.
• Neuroticism Impact: Higher Neuroticism levels systematically amplified fear, accelerated panic onset, increased the frequency of trampling and falls, and reduced overall evacuation efficiency and safety outcomes.

Significance and Claims
The paper claims that the proposed framework provides a more realistic representation of human behavior in emergency evacuations by systematically linking internal states (cognition, emotion, personality) to observable crowd dynamics. It demonstrates that incorporating these psychological factors is not merely descriptive but fundamentally alters evacuation outcomes, leading to reduced efficiency and the emergence of complex phenomena such as delays, confusion, and physical injuries.

The authors emphasize that the framework enables the systematic investigation of interactions between cognition, emotion, personality, and crowd dynamics. They acknowledge limitations, noting that the current model does not explicitly capture pro-social behaviors (e.g., helping strangers) or the full spectrum of the OCEAN personality model, and that the quantitative outcomes depend on parameter calibration. However, the qualitative trends observed across experiments support the validity of the modeling approach in capturing the heterogeneity and complexity of real-world emergency responses. The work positions itself as an incremental step toward more realistic evacuation modeling, bridging the gap between simplified physical models and the complex reality of human behavior under stress.