Adaptive Evacuation Guidance with Reinforcement Learning

Adaptive Evacuation Guidance with Reinforcement Learning

This research investigates the use of reinforcement learning to provide adaptive indoor evacuation guidance under dynamic congestion and unexpected route disruptions. A pedestrian simulation environment was developed with a Proximal Policy Optimisation (PPO) agent trained to dynamically adjust evacuation routes in response to changing conditions. Experimental results demonstrate a 17.3% reduction in average evacuation time compared with the best fixed-routing baseline, highlighting the potential of reinforcement learning for more responsive and efficient evacuation management.

Duc Tran

Artificial intelligence, AI, Machine learning, Deep Learning, reinforcement learning, Computer Science, simulation, Modelling, Optimization, Optimal Control, Operational Research, Decision-making, congestion, neural networks, Python, Evacuation, Emergency Evacuation, Adaptive Evacuation Guidance, Pedestrian Simulation, Crowd Simulation, Crowd Dynamics, Crowd Management, Dynamic Routing, Evacuation Routing, Route Blockages, Dynamic Congestion, Proximal Policy Optimisation, PPO, JuPedSim, Gymnasium, Multi-Agent Simulation, Public Safety