Evacuaidi: An AI-Driven, Causal-Informed Framework for Probabilistic and Disability-Inclusive Evacuation Guidance

When emergencies such as fires or natural disasters occur in large buildings, it is crucial that people evacuate quickly and safely. Today, new technologies like artificial intelligence (AI) can help guide people during these emergencies by suggesting the best evacuation routes. However, not everyone will always follow these instructions, especially when there is confusion, fear, or a lack of trust in the system. This is particularly important for individuals with disabilities, who may have different needs or challenges during evacuation. This dissertation focuses on designing smarter and safer evacuation systems that work well even when not everyone follows directions. It uses real data from evacuation drills at a university building, including participants with and without disabilities, to study how people move and respond to AI-based guidance systems. A new computer model was created to simulate how people behave in different emergency situations and how changes in trust or building occupancy affect the overall safety. The model also uses advanced statistics to account for uncertainty, which helps decision-makers see how safe or risky different scenarios could be. The results show that systems need to be flexible and consider human behavior in all its complexity. For example, increasing trust in the system can make evacuations faster and safer, but the benefits eventually level off. By understanding how much improvement is possible and when those improvements slow down, building managers and emergency planners can make better decisions about where to invest time and money. This research offers a new approach to designing inclusive evacuation plans that are both realistic and adaptable. It also highlights the importance of combining technology with a deep understanding of human behavior to protect everyone, especially those who are often overlooked in emergency planning.