This project tackles inefficiencies in traffic light synchronization, which disrupt traffic flow and cause unnecessary delays. Traditional sensor-based systems at intersections fail to consider the larger traffic network. To address this, the proposed solution employs a decentralized system with IoT devices and a Federated Learning Framework.
The IoT devices will function within a secure, self-contained network, communicating only with each other and the learning algorithm. This setup enables seamless peer-to-peer data sharing and processing, free from external interference. By leveraging inflow-outflow equations and real-time traffic data, the system will dynamically adjust traffic signals to improve urban traffic flow over time.
Since this system relies on peer-to-peer networking, various federated network architectures will be explored. The project will focus on developing mathematical proofs within CSC 431 to support the concept. A proof of concept plan will outline implementation details, required resources, and expected challenges. While a physical implementation is unlikely within the project’s timeline, an open-source simulation tool will provide valuable insights into traffic light behavior.
Initial research and literature review on decentralized systems, IoT devices, and Federated Learning Frameworks. Begin drafting the project proposal and identifying key components.
Develop a detailed plan for decentralized network architecture. Start working on mathematical proofs for traffic optimization and explore an open-source traffic light synchronization project.
Refine the network architecture and mathematical proofs. Conduct simulations and experiments to validate models, document findings, and adjust the project plan as necessary.
Finalize the proof of concept, prepare implementation documentation, and outline challenges. Begin drafting the final project report.
Complete the final report, including all findings, proofs, and implementation details. Prepare a presentation summarizing the project and submit all deliverables.
Proposal Drafting: Began drafting the project proposal, clearly outlining the project’s objectives and key components.
Download PDFBiweekly Update Summary (February 7 – February 21, 2025):
Research Initiated: Conducted an extensive literature review focusing on decentralized systems, IoT devices, and federated learning frameworks.
Papers read:
Component Identification: Identified critical elements for the project, including the design of a secure decentralized network, the integration of a federated learning algorithm, and the use of simulation tools for traffic light optimization.
Next Steps: Planning to further develop the network architecture and initiate preliminary work on mathematical models and simulations.
Download PDFBiweekly Update Summary (February 22 – March 7, 2025):
Pivoted project scope slightly to prioritise network architecure, design, and development - notes on this here
Read several papers on gossip-based protocols
Now considering Kademlia type of architecture so spent some time reading papers and the RS2 docs (shoutout to Emily Martins)
Managed to set up OMNeT++ simulation of downtown Victoria - check it out below
Download PDFBiweekly Update Summary (March 8 – March 21, 2025):
Installed the project and software on a Windows machine due to simulation difficulties on my Mac.
Retrieved traffic data from the City of Victoria Open Data Portal and overlaid this data on an OpenStreetMap (OSM) of downtown Victoria.
Created small test networks in NetEdit (from SUMO) to begin testing simulation topologies.
Explored optimization methods using matrices for traffic flow, modeling intersections as nodes and streets as directed edges, leading to the identification of a "standard matrix" for traffic flow in a network of intersections.
Encountered challenges with the City of Victoria’s traffic data, notably the lack of intersection-specific data and directional flow information, but identified potential methods to estimate and work with the available data, including approximations and assumptions.
Successfully got a simulation of downtown Victoria running in SUMO, using Python scripts to generate the network and visualizing traffic patterns.
Began formulating a federated learning (FL) network architecture, which evolved into a hierarchical yet partially meshed network topology with local cliques and inter-clique connections, with nodes acting as gateway nodes for communication.
Outlined a potential approach for building an FL network from scratch in OMNeT++, focusing on a standalone proof of concept before integrating SUMO via TraCI to simulate traffic mobility.
Designed a preliminary gossip protocol for federated learning in OMNeT++, highlighting local and global gossiping dynamics and how nodes could dynamically adjust their gossip weight based on traffic volume and redundancy.
For a more in-depth look into my thought process, issues, etc. check out the project log sheet here
Download PDFBiweekly Update Summary (March 22 – April 11, 2025):
Finalized the proof of concept and prepared implementation documentation.
Worked on the final project report, including all findings, proofs, and implementation details.
Prepared a presentation summarizing the project and submitted all deliverables.
Download final paper PDFThe GitHub repository for this project can be found here.