TMS2 - RL Traffic Management Models
RL Traffic Signal Controller
Deep Q-Network (DQN) based agents for adaptive traffic signal control.
Variants:
- v2: Baseline stable model optimized for throughput
- v3: Eco-friendly model with emissions optimization (14% CO2 reduction)
- v4: Challenging scenarios with incidents and demand spikes
- v5/final: Curriculum learning for generalization
Architecture:
- Dueling Double DQN with soft target updates
- State space: 10-12 dimensions (queues, phase, time, scenarios)
- Action space: 2 (keep phase / change phase)
Model Description
These models are part of the Traffic Management System 2 (TMS2) project, an intelligent traffic control system using deep learning and reinforcement learning.
Training Details
- Framework: PyTorch
- Training Platform: Google Colab (T4 GPU)
- Training Date: December 2025
Usage
import torch
# Load model
model = torch.load('model.pt')
model.eval()
# Inference
with torch.no_grad():
output = model(input_tensor)
License
Apache 2.0