Project Goal: Target 15-20% reduction in urban traffic congestion for the Athlone "Orange Loop" using Reinforcement Learning, measured by:
- Average Travel Time (ATT): -15%
- Mean Queue Length (MQL): -20%
- Data Integrity: TLS 1.3 secured telemetry pipeline
- API Gateway — REST API service on Render
- Inference Service — RL model inference on Render
- LSTM Predictor Service - LSTM Predictor Service on Render
- API Gateway Swagger UI — Interactive API docs
- Inference Service Swagger UI — Interactive API docs
- LSTM Predictor Service Swagger UI — Interactive API docs
- Mappo Simulation v2 - Mappo Traffic Flow Simulation v2 demo
- Detailed MkDocs Documentation - Complete mkdocs site
- System Architecture — Detailed system design
- API Endpoints — Quick reference and full API specifications
- API Gateway - Detailed Java API Gateway documentation
- RL Inference - Detailed RL Inference service documentation
- LSTM Predictor - Detailed LSTM Predictor service documentation
- Quick Start Guide — Get up and running quickly
- Gateway JWT Authentication Guide — JWT configuration and usage
🚀Running the System
You can start all three microservices (API Gateway, RL Inference, and LSTM Predictor) using the provided startup scripts:
start.bat./start.sh🛠️ API Setup — Docker, Environment Variables, Components, Examples, Monitoring
📖 API Reference — Quick Reference & Detailed Specifications
💬 Support — Contact, Contributing, Issues
┌─────────────────────────────────┐
│ CLIENT LAYER │
│ Web • Mobile • External API │
└────────────────┬────────────────┘
│
│ HTTPS (TLS 1.3)
│
┌──────────────────▼──────────────────┐
│ JAVA API GATEWAY (Port 8080) │
│ Spring Boot 3.2.3 │
├─────────────────────────────────────┤
│ • JWT Authentication (HS256) │
│ • Request Validation │
│ • Fallback Logic (RED signal) │
│ • Circuit Breaker Pattern │
└──────────────────┬──────────────────┘
│
┌───────────┴─────────────┐
│ │
┌───────────▼────────┐ ┌────────────▼──────────┐
│ HEALTH CHECK │ │ PREDICTION REQUESTS │
│ /health │ │ /traffic/action │
└────────────────────┘ └┬──────────────────────┘
│
┌───────────▼───────────┐
│ INFERENCE SERVICE │
│ (Port 8000) │
│ Python/FastAPI │
├───────────────────────┤
│ • MAPPO RL Model │
│ • 5-Junction Support │
│ • GRU State Mgmt │
│ • Action Masking │
└───────────┬───────────┘
│
┌─────────────┴───────────┐
│ │
┌───────────▼─────────┐ ┌───────────▼───────────┐
│ LSTM PREDICTOR │ │ ACTION SELECTION │
│ (Port 8001) │ │ (MAPPO Output) │
│ Python/FastAPI │ │ │
├─────────────────────┤ │ Actions: │
│ • Time-series LSTM │ │ 0: RED │
│ • 15-min Forecast │ │ 1: YELLOW │
│ • MAE < 10% │ │ 2: GREEN │
│ • Data Pipeline │ │ 3: GREEN_EXTENDED │
└─────────────────────┘ └───────────┬───────────┘
│
│ Signal State
│
┌───────────▼───────────┐
│ RESPONSE TO CLIENT │
│ { │
│ action: 0-3, │
│ signalState: "RED", │
│ confidence: 0.87 │
│ } │
└───────────────────────┘
This repository implements a Cloud-Native Microservices Pipeline designed for the Athlone "Orange Loop" case study.
-
Traffic Monitoring Gateway (Java/Spring Boot): Manages secure telemetry ingestion and orchestrates service communication.
- JWT authentication: Stateless token-based security with HS256 signing
- Exception handling: Dedicated exception package for RL service communication errors
- See Java API Gateway README for configuration and usage examples
-
RL Inference Service (Python/FastAPI): Hosts a trained MAPPO (Multi-Agent Proximal Policy Optimization) model to predict optimal signal timings based on real-time traffic observations from 5 junctions
-
LSTM Predictor Service (Python/FastAPI): Forecasts vehicle flow 15 minutes ahead using historical traffic patterns from SUMO simulations, targeting MAE < 10% accuracy
-
Simulation Layer (SUMO): Integrated high-fidelity traffic simulation environment for testing adaptive signal logic against baseline fixed-time controllers. YouTube Demo.
This system is specifically modeled to address the saturation flow rates and signal-timing patterns of the Athlone 'Orange Loop' corridor, providing a scalable template for Smart City traffic management in regional Irish hubs.
Two FastAPI microservices:
RL Inference Service (Port 8000)
- MAPPO agent for signal prediction
- Full docs
LSTM Predictor Service (Port 8001)
- LSTM for traffic density forecasting
- Full docs
Start both services in separate terminals:
cd rl-inference-service
python -m uvicorn app.main:app --host 0.0.0.0 --port 8000 --reloadcd lstm-predictor-service
python -m uvicorn app.main:app --host 0.0.0.0 --port 8001 --reloadTest integration:
python lstm-predictor-service/test_rl_integration.pyTUS-26-ETP-AI-Traffic-Optimisation/
├── java-api-gateway/ # Spring Boot Edge Service (JWT Auth, LSTM/RL Orchestration)
├── rl-inference-service/ # Python FastAPI (MAPPO Multi-Agent RL Inference)
├── lstm-predictor-service/ # Python FastAPI (Time-Series Traffic Density Forecasting)
├── SUMO/ # Traffic Simulation (Athlone 'Orange Loop' Network & Results)
├── docs/ # Central Documentation (MkDocs source & guides)
├── docker-compose.yml # Container orchestration for all services
├── start.bat / start.sh # Multi-platform quickstart scripts
└── README.md # This file
For a detailed breakdown of every file, please refer to our detailed Project Structure documentation.
2026 • Edgars Peskaitis • Joe O'Regan • David Claffey • Adam O Neill Mc Knight