SUMO-TSC-evaluation

This repo includes the source data & code for our paper, "Evaluation of Traffic Signal Control at Varying Demand Levels: A Comparative Study", in IEEE ITSC 2023.

The code structure is based on RESCO Benchmark. We have modified/added functionalities for our paper use.

Structure

• pfrl is a local package of pfrl modified for model testing purposes.
• resco_benchmark is the modified SUMO-based traffic signal control package with various useful built-in functionalities. We make modifications as follows:
  • agent_tf2.0: we convert all tensorflow 1.x uses to a tf2.x-compatible version.
  • Scenario: We modified the original Ingolstadt scenario to make it work better with TSC algorithms. Besides, we fixed some map inconsistencies in signal_config.py.
  • Demands: we created 3 static and 1 time-varying demand files for our evaluation. They are named as ingolstadt7low, ingolstadt7mid, ingolstadt7hig (static) and ingolstadt7x (dynamic).
  • Output: Vehicle data is retrieved as output from SUMO config. We enabled the retrieval of vehicle (trip) data and unfinished trips.
• results includes all training and testing results from our experiments, in which pace_plotting.py, training_plotting.py, vehicle_info.py are three visualization scripts. xml_processing.py and csv_processing_ing7.py are postprocessing scripts for SUMO output data.

Usage

For algorithm training and testing, run resco_benchmark/main.py with corresponding parameters. For output analysis and visualizations, use the scripts in results/.

*Results are kept in Vanderbilt University Institutional Repository (link). Unzip "data.zip" in 'results/' for processing.

Contact

• Author: Zhiyao Zhang, Marcos Quinones-Grueiro, William Barbour, Yuhang Zhang, Gautam Biswas, and Daniel Work
• Affiliation: Institute for Software Integrated Systems, Vanderbilt University
• First-author email: zhiyao.zhang@vanderbilt.edu