Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Human-Centric Traffic Signal Control for Equity: A Multi-Agent Action Branching Deep Reinforcement Learning Approach

Xiaocai Zhang, Neema Nassir, Lok Sang Chan, Milad Haghani

TL;DR

This work tackles fairness in multimodal traffic signal control by addressing traveler-level equity across vehicles, pedestrians, and transit. It introduces MA2B-DDQN, a human-centric multi-agent DRL framework with an action-branching discrete control architecture that couples locally optimized phase splits with a global duration decision, trained centrally and executed decentrally. A custom, multimodal reward penalizes delays across all traveler types, and the method is validated on seven Melbourne corridor scenarios, where MA2B-DDQN consistently achieves lower total-delayed-travelers and robust performance relative to diverse baselines. The findings demonstrate the practicality of scalable, equity-aware TSC with discrete action spaces and underline potential extensions to multi-objective optimization and emergency-priority integration for real-world deployment.

Abstract

Coordinating traffic signals along multimodal corridors is challenging because many multi-agent deep reinforcement learning (DRL) approaches remain vehicle-centric and struggle with high-dimensional discrete action spaces. We propose MA2B-DDQN, a human-centric multi-agent action-branching double Deep Q-Network (DQN) framework that explicitly optimizes traveler-level equity. Our key contribution is an action-branching discrete control formulation that decomposes corridor control into (i) local, per-intersection actions that allocate green time between the next two phases and (ii) a single global action that selects the total duration of those phases. This decomposition enables scalable coordination under discrete control while reducing the effective complexity of joint decision-making. We also design a human-centric reward that penalizes the number of delayed individuals in the corridor, accounting for pedestrians, vehicle occupants, and transit passengers. Extensive evaluations across seven realistic traffic scenarios in Melbourne, Australia, demonstrate that our approach significantly reduces the number of impacted travelers, outperforming existing DRL and baseline methods. Experiments confirm the robustness of our model, showing minimal variance across diverse settings. This framework not only advocates for a fairer traffic signal system but also provides a scalable solution adaptable to varied urban traffic conditions.

Human-Centric Traffic Signal Control for Equity: A Multi-Agent Action Branching Deep Reinforcement Learning Approach

TL;DR

This work tackles fairness in multimodal traffic signal control by addressing traveler-level equity across vehicles, pedestrians, and transit. It introduces MA2B-DDQN, a human-centric multi-agent DRL framework with an action-branching discrete control architecture that couples locally optimized phase splits with a global duration decision, trained centrally and executed decentrally. A custom, multimodal reward penalizes delays across all traveler types, and the method is validated on seven Melbourne corridor scenarios, where MA2B-DDQN consistently achieves lower total-delayed-travelers and robust performance relative to diverse baselines. The findings demonstrate the practicality of scalable, equity-aware TSC with discrete action spaces and underline potential extensions to multi-objective optimization and emergency-priority integration for real-world deployment.

Abstract

Coordinating traffic signals along multimodal corridors is challenging because many multi-agent deep reinforcement learning (DRL) approaches remain vehicle-centric and struggle with high-dimensional discrete action spaces. We propose MA2B-DDQN, a human-centric multi-agent action-branching double Deep Q-Network (DQN) framework that explicitly optimizes traveler-level equity. Our key contribution is an action-branching discrete control formulation that decomposes corridor control into (i) local, per-intersection actions that allocate green time between the next two phases and (ii) a single global action that selects the total duration of those phases. This decomposition enables scalable coordination under discrete control while reducing the effective complexity of joint decision-making. We also design a human-centric reward that penalizes the number of delayed individuals in the corridor, accounting for pedestrians, vehicle occupants, and transit passengers. Extensive evaluations across seven realistic traffic scenarios in Melbourne, Australia, demonstrate that our approach significantly reduces the number of impacted travelers, outperforming existing DRL and baseline methods. Experiments confirm the robustness of our model, showing minimal variance across diverse settings. This framework not only advocates for a fairer traffic signal system but also provides a scalable solution adaptable to varied urban traffic conditions.
Paper Structure (29 sections, 15 equations, 25 figures, 16 tables)

This paper contains 29 sections, 15 equations, 25 figures, 16 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Traffic simulation environment
  4. Reward mechanism
  5. Multi-agent DRL framework
  6. Value-based DRL methods
  7. Policy-based DRL methods
  8. Hybrid DRL methods
  9. Summary
  10. Methodology
  11. Overview
  12. Environment and state
  13. Agent and action
  14. Reward mechanism
  15. Network architecture
  16. ...and 14 more sections

Figures (25)

  • Figure 1: Conceptual overview of human-centric multimodal traffic signal control: travelers include private-vehicle occupants, public-transit passengers, cyclists, and pedestrians, with advanced sensing/AI enabling real-time multimodal user data acquisition and pre-processing.
  • Figure 2: Workflow of the proposed MA2B-DDQN for human-centric multi-agent traffic signal control.
  • Figure 3: The corridor selected for experiments features three intersections and is equipped with 8 entry/exit zones (indexed from 1 to 8) for vehicle simulations. Additionally, there are 12 entry/exit zones (indexed from 9 to 20) for pedestrian simulations, which do not account for the direction of pedestrian movements.
  • Figure 4: Phase and stage diagram of the traffic signal system.
  • Figure 5: Illustration of the action space via the action branching architecture.
  • ...and 20 more figures