Icon Legend

This session is not in your schedule.

This session is in your schedule. Click again to remove it.

Back

7 Views

ICTD 2025

Artificial Intelligence (AI) in Transportation

POSTER RECEPTION (View list of posters here)

A Novel Co-optimization Framework for Autonomous Driving Simulation in Large-scale Urban Scenario

Tuesday, June 10, 2025

4:45 PM – 6:00 PM MT

Primary Author(s)

Jiancong Chen

Reasearch Assistant
PURDUE UNIVERSITY - WEST LAFAYETTE
West Lafayette, IN, United States

Presenter(s)

Jiancong Chen

Reasearch Assistant
PURDUE UNIVERSITY - WEST LAFAYETTE
West Lafayette, IN, United States

Co-Author(s)

Yansong Qu

Teaching Assistant
PURDUE UNIVERSITY - WEST LAFAYETTE
Lafayette, IN, United States
Zihao Sheng (he/him/his)

Reasearch Assistant
University of Wisconsin–Madison
Madison, WI, United States
Zilin Huang, n/a

Reasearch Assistant
University of Wisconsin–Madison
Madison, Wisconsin, United States
YC

Yen-Jung Chen, n/a

Master's Student
PURDUE UNIVERSITY - WEST LAFAYETTE
West Lafayette, Indiana, United States
Sikai Chen, n/a

Assistant Professor
University of Wisconsin-Madison
Madison, Wisconsin, United States
Samuel Labi, n/a

Professor
Purdue University
West Lafayette, Indiana, United States

Submitter(s)

Jiancong Chen

Reasearch Assistant
PURDUE UNIVERSITY - WEST LAFAYETTE
West Lafayette, IN, United States

Abstract: Growing interest in autonomous driving technologies continues to raise serious concerns about the efficacy of their solutions in complex urban networks. Most studies in this domain have developed solutions based on a single-vehicle simulation platform. As such, these studies have analyzed autonomous vehicle operations from a limited perspective. For this reason, it is difficult to generalize their solutions to scenarios involving large-scale and complex urban road networks, and dense traffic. In this research, we present a novel co-optimization framework to achieve macro-micro perspective autonomous driving simulation. Specifically, our framework first uses pre-collected traffic data to train a Simulation of Urban Mobility (SUMO)-based macrosimulation module that forecasts traffic states at specific time and route sets in a customized traffic network. Then, we integrate the output spatial-temporal features and traffic flow characteristics into a CARLA-based microsimulation module, to serve as prior knowledge for optimizing autonomous driving decision making. To remove the human-driven intervention, we introduce a deep reinforcement learning model that learns the driving strategies guided by the prior knowledge and the specifically designed reward functions. A numerical experiment for the proposed framework produced results that suggest that the simulation process can simultaneously focus on the driving state of individual vehicles as well as the dynamic changes of the road traffic network. This is particularly advantageous in scenarios characterized by large and complex road networks in urban areas.

Learning Objectives:

Attendees can expect to learn the following from this session:

Establish and verify the co-optimization framework for autonomous driving simulation in large-scale urban scenario.
Perform comprehensive experiments with some typical autonomous driving benchmarks.
Apply the macro-micro perspective simulation method to other tasks in the field of autonomous driving.