UrbanMARL Documentation

UrbanMARL is a scalable, GPU-vectorized multi-agent reinforcement learning (MARL) framework and digital twin platform for 6G geospatial radio environments, Unmanned Aerial Vehicle (UAV) swarms, Mobile Edge Computing (MEC) networks, and physical layer security.

Built natively on TorchRL, PyTorch, TensorDict, and BenchMARL, UrbanMARL vectorizes 3D spatial ray-casting, ITU-R P.1410 urban map procedural generation, mmWave radio propagation, kinematic mobility models, and M/M/c queuing dynamics across hundreds of concurrent environments.

Documentation Navigation:

• Getting Started
  • Installation Guide
  • Quickstart Tutorial
• Digital Twins Architecture
  • Geospatial 3D Urban Map Digital Twin
  • Rotary-Wing Aerodynamics Digital Twin
  • 360° LiDAR Rangefinder Digital Twin
  • Kinematic & Mobility Digital Twin
  • Radio Network Digital Twin
  • MEC Queue Digital Twin
• Agents & Scenarios Architecture
  • Agent Taxonomy & State Spaces
  • Urban Scenario Benchmarks
  • BenchMARL Integration Guide
• Security & Privacy Architecture
  • Physical Layer Security & Privacy
• API Reference
  • Environments API
  • Models API
  • Scenarios API
  • Callback API

Project Modules & Features

• Geospatial 3D Digital Twin: Procedural ITU-R P.1410 urban terrain parameterized by building coverage (\(\alpha\)), density (\(\beta\)), and height distribution (\(\gamma\)). Includes GPU-accelerated ray-casting for Line-of-Sight (LoS) and building collision detection.

• Radio Network Digital Twin: Vectorized 29 GHz mmWave channel modeling Friis path loss, LoS/NLoS attenuation shifts, interference, SINR, and Shannon channel capacity.

• MEC Queue Digital Twin: Pure PyTorch M/M/c queuing system computing server utilization, queue lengths, task waiting times, and offloading execution delays.

• Kinematic & Mobility Twin: 3D motion models for UAVs and dynamic ground User Equipments (UEs).

• Physical Layer & Network Security: Frameworks for physical layer security (PLS), anti-jamming, adversarial agent detection, and privacy-preserving task offloading.

• BenchMARL & TorchRL Integration: Native support for BenchMARL task APIs (UrbanEnvTask). Compatible with SOTA MARL algorithms (MAPPO, MADDPG, MASAC, IPPO, IDDPG, ISAC, QMIX).

Indices and Tables

• Index

• Module Index

• Search Page