Technology & Architecture Stack
A comprehensive architectural disclosure for municipal Chief Technology Officers, transportation engineers, and systems architects. Transparently distinguishing implemented simulation code from active hardware development.
Python Scientific & AI Core
Python serves as the backbone for algorithmic modeling, TraCI socket communication, simulation orchestration, and data transformation pipelines.
Key Libraries
PyTorch, NumPy, SciPy, Stable-Baselines3, Scikit-learn, Ray Tune.
Operational Role
Executes RL policy evaluation, state feature extraction, and mathematical queue optimization.
Eclipse SUMO & TraCI Simulation Engine
Eclipse SUMO (Simulation of Urban MObility) is an open-source, highly portable, microscopic vehicular traffic simulator. Connected via the TraCI (Traffic Control Interface) socket protocol, it executes sub-second physical simulation runs with realistic car-following dynamics (Krauss / IDM models).
Simulates individual vehicle speed, braking, lane changing, and acceleration curves.
Allows Python agents to inspect queues and modify signal phases step-by-step.
Loads synthetic and real-world origin-destination matrices from municipal count logs.
Geospatial Engine: PostGIS, GeoPandas & OpenStreetMap
Spatial intelligence is powered by OpenStreetMap topology converted into PostGIS network graphs. GeoPandas and Shapely execute high-speed spatial joins, corridor bounding boxes, and Low-Emission Zone polygon geofences.
PostGIS 3.4 on PostgreSQL 16
Manages street centerlines, junction coordinates, crosswalk geometries, and spatial indexing (`GIST`) for microsecond proximity queries across metropolitan territories.
OSMnx & OpenStreetMap Ingestion
Downloads, models, and visualizes street networks directly from OSM, generating navigable directed multigraphs with edge speeds, lane counts, and turn restrictions.
Machine Learning: DQN, PPO, LSTM & GRU
Our dual AI pipeline combines predictive sequence modeling with closed-loop policy optimization:
Agents learn to dynamically adjust green phase durations by observing queue lengths, wait time variances, and downstream discharge capacities.
Recurrent neural networks predict flow surges across arterial links up to 60 minutes in advance, enabling the signal controllers to adjust offsets before queues propagate.
Edge Computer Vision: YOLOv8 & RTSP Pipeline
Processes municipal RTSP video streams at the network edge. Tracks and classifies vehicles into classes (cars, buses, delivery trucks, motorcycles) while counting lane queue densities.
Backend & Cloud Architecture: FastAPI & Azure AKS
High-concurrency asynchronous API layer engineered in FastAPI and Python 3.11 with containerized orchestration on Microsoft Azure Kubernetes Service (AKS).
Sub-millisecond serialization of municipal telemetry feeds using Pydantic V2.
Maintains active intersection signal states, current phases, and operator overrides.
Standard protocol bridging legacy 170/2070/ATC signal cabinets with cloud telemetry.