Page 18 - 2024S

UEC Int’l Mini-Conference No.52 11

Optimizing Pedestrian Crossings and Traffic Signal Control for Urban

Mobility

Massey VO ∗1 and Takeo FUJII 2

1 UEC Exchange Study Program (JUSST Program)
2 Advanced Wireless Communication Research Center (AWCC)
The University of Electro-Communications, Tokyo, Japan

Abstract

This research aims to analyze the impact of pedestrian crossing timings on vehicular traffic flow
in urban environments and propose an adaptive traffic signal control algorithm using Q-learning.
The operational and safety performance of the proposed algorithm was tested using the Simulation
of Urban Mobility (SUMO). The results demonstrate significant improvements in traffic flow and
safety, especially under high Market Penetration Rates (MPRs) of Connected and Automated Vehicles
(CAVs).
Keywords: Urban Traffic, Q-learning, Traffic Signal Control, SUMO, Adaptive Algorithms, Con-
nected and Automated Vehicles (CAVs)

1 Introduction gorithm that dynamically adjusts signal timings
to improve both vehicular traffic flow and pedes-
Urban traffic congestion, especially during large trian safety at urban intersections. The inte-
events like the 2020 Tokyo Olympics, highlights gration of Connected and Automated Vehicles
the need for more efficient traffic management (CAVs) data enhances the algorithm’s accuracy
solutions. Enhancing road safety and promoting and responsiveness, making it a promising solu-
efficient urban mobility by reducing idling times tion for future smart city applications.
at traffic lights are critical goals of this research.
Traffic signal control is a crucial element in
managing urban traffic flow. Traditional meth-
ods, such as Fixed-Time (FT) control, rely 2 Methodology
on pre-set timers that do not adapt to real-
time traffic conditions, leading to inefficiencies. The methodology section outlines the design
Adaptive Traffic Control Systems (ATCS), such and implementation of the proposed Q-learning-
as SCATS and SCOOT, adjust signal timings based traffic signal control algorithm.
based on real-time data from sensors but re-
quire substantial infrastructure. Recent ad-
vances in machine learning, particularly rein-
2.1 State Representation
forcement learning algorithms like Q-learning,
offer new possibilities for optimizing traffic sig- In the Q-learning algorithm, states are defined
nal control by learning from real-time traffic
conditions. based on the queue lengths of each approach at
the intersection. The state representation is cru-
This study focuses on developing a Q-
cial for the algorithm to understand the traffic
learning-based adaptive traffic signal control al-
conditions and make informed decisions. Table
∗ The author is supported by JASSO Scholarship. 1 presents the state set used in this study.