Tutorial #1: Multi Agent Systems for Emergency Response

Summary form only given. The complete presentation was not made available for publication as part of the conference proceedings. Emergency response to incidents such as accidents, crimes, and wildfires is a major problem faced by communities. Emergency response management (ERM) comprises several stages and sub-problems like forecasting, detection, allocation, and dispatch. The design of principled approaches to tackle each problem is necessary to create efficient ERM pipelines. This talk will go through the design of principled decision-theoretic and data-driven approaches to tackle emergency incidents. It will discuss the data collection, cleansing, and aggregation as well as some models and methods we used to solve an imbalanced classification problem. Further, we will explain how large multi-agent systems can be used to tackle emergency scenarios under dynamic environments and communication and state uncertainty. We will go through fundamental modeling paradigms like Markov decision processes, semi-Markov decision processes, and partially-observable Markov decision processes and how promising actions can be found for stochastic control problems. As case studies, we will specifically look at emergency incidents like wildfires and road accidents. We will also go through two open-source datasets that we have created for the research community to use regarding traffic accidents and wildfires.