Supervisors

Suitable for

Abstract

Large scale mass assemblies such as the FIFA world cup bear a potential risk of disaster, either natural or man-made, resulting in crisis situations necessitating emergency evacuation. Such emergency situations can be complex, requiring large crowds to be navigated through a limited number of narrow passages quickly in a safe fashion. Therefore, in order for large infrastructures to be evacuated effectively, a reliable evacuation strategy must be developed, evaluated and implemented. Agent-based models, which simulate complex systems in a bottom-up fashion by explicitly modelling individuals and their decision-making processes, provide environments to evaluate the practical effectiveness of evacuation strategies. This project focusses on the further development and specialisation of agent-based models for emergency evacuation. In short, the main goals of this project are: A.     Formalising crowd modelling complexities using agent-based approaches, including: a.       Realistically modelling a synthetic population of individuals. b.       Hierarchical crowd structures and crowd population dynamics. c.       Behaviour modelling: pedestrian movements, mobility patterns, stampede physics, interactions, interdependence, and interruptions; co-operation and co-ordination. B.      Developing a crowd simulation and analysis framework to: a.       Model programable spatially explicit environments and building structures at real scale (see Fig 1). b.       Conceptualize mass gathering configurations (events, routines, mobility patterns). c.       Specify and evaluate emergency evacuation scenarios. d.       Propose and test new evacuation strategies / algorithms. e.       Use optimization algorithms for comparing different strategies and building evacuation plans. C.      Explore techniques to enhance simulation speed and scale using modern techniques: a.       Adopting a network-centric design based on message-passing in order to exploit highly parallelised hardware and sparse tensor calculus for fast simulation. b.       Integrating agent-based models with deep learning architectures for fast calibration of model parameters via gradient-based methods. References & Recommended Reading: [1] Imran Mahmood, Muhammad Haris, and Hessam Sarjoughian. 2017. Analyzing Emergency Evacuation Strategies for Mass Gatherings using Crowd Simulation And Analysis framework: Hajj Scenario. In Proceedings of the 2017 ACM SIGSIM Conference on Principles of Advanced Discrete Simulation (SIGSIM-PADS '17). Association for Computing Machinery, New York, NY, USA, 231–240. https://doi.org/10.1145/3064911.3064924 [2] I. Mahmood, T. Nadeem, F. Bibi and X. Hu, "Analyzing Emergency Evacuation Strategies For Large Buildings Using Crowd Simulation Framework," 2019 Winter Simulation Conference (WSC), 2019, pp. 3116-3127, doi: 10.1109/WSC40007.2019.9004906. [3] S. Long, D. Zhang, S. Li, S. Yang and B. Zhang, "Simulation-Based Model of Emergency Evacuation Guidance in the Metro Stations of China," in IEEE Access, vol. 8, pp. 62670-62688, 2020, doi: 10.1109/ACCESS.2020.2983441. [4] Pidd, M., F. N. De Silva, and R. W. Eglese. "A simulation model for emergency evacuation." European Journal of operational research 90, no. 3 (1996): 413-419. [5] Tan, Lu, Mingyuan Hu, and Hui Lin. "Agent-based simulation of building evacuation: Combining human behavior with predictable spatial accessibility in a fire emergency." Information Sciences 295 (2015): 53-66. [6] Li, Yang, Maoyin Chen, Zhan Dou, Xiaoping Zheng, Yuan Cheng, and Ahmed Mebarki. "A review of cellular automata models for crowd evacuation." Physica A: Statistical Mechanics and its Applications 526 (2019): 120752

[7] Graph Representation Learning 2022-2023, Department of Computer Science Lecture Ismail Ilkan Ceylan https://www.cs.ox.ac.uk/teaching/courses/2022-2023/grl/ [8] Hamilton, William L. "Graph representation learning." Synthesis Lectures on Artifical Intelligence and Machine Learning 14.3 (2020): 1-159 https://www.cs.mcgill.ca/~wlh/grl_book/files/GRL_Book.pdf [9] Macal, Charles M. "Everything you need to know about agent-based modelling and simulation." Journal of Simulation 10, no. 2 (2016): 144-156 [10] Abar, Sameera, Georgios K. Theodoropoulos, Pierre Lemarinier, and Gregory MP O’Hare. "Agent Based Modelling and Simulation tools: A review of the state-of-art software." Computer Science Review 24 (2017): 13-33.

Required Prerequisites: Familiarity with Python, Artificial Intelligence. Desirable Prerequisites: Familiarity with PyTorch, Graph Representation Learning, Machine Learning, Bayesian Statistical Probabilistic Programming, Combinatorial Optimisation, Computational Game Theory.