United States Department of Transportation
i
Multi-Agent Reinforcement Learning-Based Pedestrian Dynamics Models for Emergency Evacuation
-
2022-06-01
[PDF-2.88 MB]
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Publication/ Report Number:
-
Resource Type:
-
Geographical Coverage:
-
Edition:Final Report 4/2019-5//2022
-
Corporate Publisher:
-
Abstract:An efficient and safe evacuation of passengers is important during emergencies. Overcapacity on a route can cause an increased evacuation time. Decision making is essential to optimally guide and distribute pedestrians to multiple routes while ensuring safety. Developing an optimal pedestrian path planning route while considering learning dynamics and uncertainties in the environment generated from pedestrian behavior is challenging. While previous evacuation planning studies have focused on either simulation of realistic behaviors or simple route planning, the best route decisions with several intermediate decision points, especially under real-time changing environments, have not been considered. This project develops an optimal navigation model providing more navigation guidance for evacuation emergencies to minimize the total evacuation time while considering the influence of other passengers based on the social-force model. The integration of the optimal navigation model was ultimately able to reduce the overall evacuation time by 10.6%, compared to the use of only one modeling approach. This study further simulated robot agents to explore and evacuate from a dynamic environment with or without collaboration using the MARL Q-learning algorithm. The multi-agent collaboration method was found to perform better than the single-agent exploration regarding the evacuation time, death counts, and reward both in the static threats and dynamic threats environment. Results were discussed, and future directions were given in the end.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type:
