Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Facilitating Emergency Vehicle Passage in Congested Urban Areas Using Multi-agent Deep Reinforcement Learning

Haoran Su

TL;DR

This work tackles EMV passage in congested urban networks by proposing two complementary MARL-based solutions: EMVLight, a decentralized framework that jointly optimizes EMV routing and traffic signal pre-emption (including emergency lanes) to reduce EMV TT by up to ~42.6% while cutting average non-EMV TT by ~23.5%; and MAPPO-DQJL, which orchestrates dynamic queue-jump lanes via centralized training with decentralized execution in mixed traffic to cut EMV TT by up to ~39.8% and reduce non-EMV lane changes by ~55.7%. A transformer-based MAPPO-DQJL architecture coordinates CAVs (and includes HDVs as learning agents) to form efficient DQJLs under variable CAV penetration and traffic densities. Separately, the dissertation presents an intersection-aware NYC EMS accessibility model that incorporates intersection density and demographic factors to identify vulnerable regions and demonstrates potential EMS accessibility improvements with EMVLight (e.g., serving 95% of residents within 4 minutes). Taken together, these contributions offer a scalable, data-driven framework for emergency vehicle mobility and EMS equity in dense urban environments, with implications for policymakers and urban planners seeking safer, more responsive transportation systems.

Abstract

Emergency Response Time (ERT) is crucial for urban safety, measuring cities' ability to handle medical, fire, and crime emergencies. In NYC, medical ERT increased 72% from 7.89 minutes in 2014 to 14.27 minutes in 2024, with half of delays due to Emergency Vehicle (EMV) travel times. Each minute's delay in stroke response costs 2 million brain cells, while cardiac arrest survival drops 7-10% per minute. This dissertation advances EMV facilitation through three contributions. First, EMVLight, a decentralized multi-agent reinforcement learning framework, integrates EMV routing with traffic signal pre-emption. It achieved 42.6% faster EMV travel times and 23.5% improvement for other vehicles. Second, the Dynamic Queue-Jump Lane system uses Multi-Agent Proximal Policy Optimization for coordinated lane-clearing in mixed autonomous and human-driven traffic, reducing EMV travel times by 40%. Third, an equity study of NYC Emergency Medical Services revealed disparities across boroughs: Staten Island faces delays due to sparse signalized intersections, while Manhattan struggles with congestion. Solutions include optimized EMS stations and improved intersection designs. These contributions enhance EMV mobility and emergency service equity, offering insights for policymakers and urban planners to develop safer, more efficient transportation systems.

Facilitating Emergency Vehicle Passage in Congested Urban Areas Using Multi-agent Deep Reinforcement Learning

TL;DR

This work tackles EMV passage in congested urban networks by proposing two complementary MARL-based solutions: EMVLight, a decentralized framework that jointly optimizes EMV routing and traffic signal pre-emption (including emergency lanes) to reduce EMV TT by up to ~42.6% while cutting average non-EMV TT by ~23.5%; and MAPPO-DQJL, which orchestrates dynamic queue-jump lanes via centralized training with decentralized execution in mixed traffic to cut EMV TT by up to ~39.8% and reduce non-EMV lane changes by ~55.7%. A transformer-based MAPPO-DQJL architecture coordinates CAVs (and includes HDVs as learning agents) to form efficient DQJLs under variable CAV penetration and traffic densities. Separately, the dissertation presents an intersection-aware NYC EMS accessibility model that incorporates intersection density and demographic factors to identify vulnerable regions and demonstrates potential EMS accessibility improvements with EMVLight (e.g., serving 95% of residents within 4 minutes). Taken together, these contributions offer a scalable, data-driven framework for emergency vehicle mobility and EMS equity in dense urban environments, with implications for policymakers and urban planners seeking safer, more responsive transportation systems.

Abstract

Emergency Response Time (ERT) is crucial for urban safety, measuring cities' ability to handle medical, fire, and crime emergencies. In NYC, medical ERT increased 72% from 7.89 minutes in 2014 to 14.27 minutes in 2024, with half of delays due to Emergency Vehicle (EMV) travel times. Each minute's delay in stroke response costs 2 million brain cells, while cardiac arrest survival drops 7-10% per minute. This dissertation advances EMV facilitation through three contributions. First, EMVLight, a decentralized multi-agent reinforcement learning framework, integrates EMV routing with traffic signal pre-emption. It achieved 42.6% faster EMV travel times and 23.5% improvement for other vehicles. Second, the Dynamic Queue-Jump Lane system uses Multi-Agent Proximal Policy Optimization for coordinated lane-clearing in mixed autonomous and human-driven traffic, reducing EMV travel times by 40%. Third, an equity study of NYC Emergency Medical Services revealed disparities across boroughs: Staten Island faces delays due to sparse signalized intersections, while Manhattan struggles with congestion. Solutions include optimized EMS stations and improved intersection designs. These contributions enhance EMV mobility and emergency service equity, offering insights for policymakers and urban planners to develop safer, more efficient transportation systems.

Paper Structure

This paper contains 151 sections, 49 equations, 43 figures, 23 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Background and Motivation
  3. Research Challenges
  4. Coupling of EMV Navigation and Traffic Signal Pre-emption
  5. Intra-link Movement for EMV
  6. Dissertation Contributions and Summary
  7. EMVLight
  8. MAPPO-DQJL
  9. EMS Accessibility Study of NYC
  10. Dissertation Outline
  11. EMVLight: a Multi-agent Reinforcement Learning Framework for an Emergency Vehicle Decentralized Routing and Traffic Signal Control System
  12. Introduction
  13. Literature Review
  14. Preliminaries
  15. Traffic map, movement and signal phase
  16. ...and 136 more sections

Figures (43)

  • Figure 1: NYC 911 End-to-End ERTs for July 2014 and July 2024, across all incident categories.
  • Figure 2: NYC 911 EMV travel times for July 2014 and July 2024, across all incident categories.
  • Figure 3: Traffic movements of a four-way two-lane bidirectional intersection.
  • Figure 4: Top: 8 signal phases; Left: phase #2 illustration; Right: phase #5 illustration.
  • Figure 5: A demonstration of an emergency lane for EMV passage. Top: the EMV has to follow the congested queue due to insufficient emergency capacity; Bottom: the EMV is traversing on the yielded emergency lane due to sufficient emergency capacity.
  • ...and 38 more figures

Theorems & Definitions (2)

  • Remark 1
  • Definition 1: Dynamic Queue Jump Lane