A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
A reinforcement learning approach to real-time emergency evacuation
In emergency situations, ensuring the fast and secure evacuation of occupants from buildings is critical. Traditional evacuation plans often rely on static routes that may not adapt well to dynamic conditions of the environment such as fallen objects or fire blocking the way. This thesis presents an innovative application of reinforcement learning (RL) to develop a dynamic evacuation route guidance system that adapts in real time to changing conditions. We employ deep Q-Networks(DQN) and proximal policy optimization (PPO) to optimize evacuation strategies based on realtime data. Our system is designed to minimize the evacuation time and increase the safety ofevacuees by dynamically adjusting routes as the environment changes. We compare our RL based system against traditional static evacuation plans in simulated environments that include varying complexities, such as different building layouts and fire spread patterns. Our results demonstrate that RL approaches can outperform static methods, particularly in scenarios with high unpredictability. This study contributes to emergency management by demonstrating the potential of machine learning to enhance safety during critical situations.
A reinforcement learning approach to real-time emergency evacuation
In emergency situations, ensuring the fast and secure evacuation of occupants from buildings is critical. Traditional evacuation plans often rely on static routes that may not adapt well to dynamic conditions of the environment such as fallen objects or fire blocking the way. This thesis presents an innovative application of reinforcement learning (RL) to develop a dynamic evacuation route guidance system that adapts in real time to changing conditions. We employ deep Q-Networks(DQN) and proximal policy optimization (PPO) to optimize evacuation strategies based on realtime data. Our system is designed to minimize the evacuation time and increase the safety ofevacuees by dynamically adjusting routes as the environment changes. We compare our RL based system against traditional static evacuation plans in simulated environments that include varying complexities, such as different building layouts and fire spread patterns. Our results demonstrate that RL approaches can outperform static methods, particularly in scenarios with high unpredictability. This study contributes to emergency management by demonstrating the potential of machine learning to enhance safety during critical situations.
A reinforcement learning approach to real-time emergency evacuation
Viola, Nicolas (author)
2024-01-01
mDA 24 012
Theses
Electronic Resource
English
Real Fire Emergency Evacuation of Disabled People
| British Library Conference Proceedings | 1993