Hlynr Intercept – Radar-Based Missile Defense Simulator

A production-ready RL environment for training interceptor missiles using realistic radar-only observations. Based on PAC-3/THAAD interceptor specifications, this system trains agents that have **no direct knowledge** of incoming threats and must rely entirely on simulated radar sensors - just like real-world missile defense systems.

Overview

Hlynr Intercept simulates authentic defensive missile interception scenarios where AI agents learn to:

• Search and acquire targets using realistic radar systems (5000m range, 60° beam width)
• Track through noise with range-dependent measurement uncertainty
• Intercept under constraints with fuel limits, thrust vectoring, and 6-DOF physics
• Handle detection failures when radar loses lock or targets move outside sensor range

The system features a complete 17-dimensional radar observation space that provides only sensor-realistic information, making trained policies directly transferable to real hardware without the "sim-to-real gap" of omniscient training environments.

Collaborators

• Roman Slack (RIT Rochester)
• Quinn Hasse (UW Madison)

Current System

🎯 Production-Ready Implementation: The rl_system/ directory contains a complete, radar-based missile defense simulator ready for training and deployment.

📡 Key Features:

• Authentic radar physics: Range limits, beam width constraints, detection failures
• PAC-3 interceptor modeling: 500kg mass, 50 m/s² acceleration, realistic fuel consumption
• 17D observation space: Radar-only with perfect self-state knowledge
• Multi-scenario training: Easy → Medium → Hard difficulty progression
• Production deployment: FastAPI server + offline batch evaluation

Repository Structure

.
├── README.md                    # Project overview
├── SYSTEM_ARCHITECTURE_REPORT.md # Technical architecture analysis  
├── rl_system/                  # 🚀 PRODUCTION SYSTEM
│   ├── README.md              # Complete usage guide
│   ├── SYSTEM_DESIGN.md       # Technical specifications
│   ├── PHYSICS_FEATURES.md    # Advanced Physics v2.0 documentation
│   ├── core.py                # 17D radar observations + safety
│   ├── environment.py         # 6DOF physics simulation
│   ├── physics_models.py      # ISA atmosphere, Mach drag, enhanced wind
│   ├── physics_randomizer.py  # Domain randomization framework
│   ├── train.py               # PPO training with adaptive features
│   ├── inference.py           # FastAPI server + offline evaluation
│   ├── logger.py              # Unified timestamped logging
│   ├── config.yaml            # Main configuration (with physics v2.0)
│   ├── scenarios/             # Easy/Medium/Hard presets
│   ├── tests/                 # Comprehensive physics validation tests
│   ├── Images/                # Documentation assets
│   └── requirements.txt       # Dependencies
├── deprecated/                # Legacy implementations
├── training/                  # Research prototypes (cluttered)
├── hlynr_bridge/             # Unity integration components
└── utilities/                # Episode generation tools

Quick Start

Navigate to the production system:

cd rl_system/

See complete documentation:

• rl_system/README.md - Usage guide with examples
• rl_system/SYSTEM_DESIGN.md - Technical specifications

Train a high-performance radar-based interceptor:

# Install dependencies
pip install -r requirements.txt

# Train with optimized configuration (5M steps, ~25-30 minutes)
python train.py --config config.yaml

# Monitor training in real-time with TensorBoard
tensorboard --logdir logs

# Access TensorBoard at: http://localhost:6006

Curriculum learning approach (recommended for best results):

# Stage 1: Train on easy scenario (1-2M steps)
python train.py --config scenarios/easy.yaml

# Stage 2: Continue training on standard config
python train.py --config config.yaml

# Stage 3: Evaluate on hard scenario
python inference.py --model checkpoints/best --mode offline --scenario hard

Deployment:

# Run inference server for real-time interception
python inference.py --model checkpoints/best --mode server

# Batch evaluation with JSON export
python inference.py --model checkpoints/best --mode offline --scenario medium



python inference.py --model checkpoints/model_400000_steps.zip --config config.yaml --mode offline

Expected training results:

• 1M steps (~5 min): 30-40% interception success rate
• 3M steps (~15 min): 60-70% interception success rate
• 5M steps (~25 min): 75-85% interception success rate

System Highlights

• 🎯 Radar-Only Training: No omniscient observations - interceptors learn to search, acquire, and track targets through realistic sensor limitations
• 🚀 PAC-3 Physics: Authentic 6DOF dynamics with thrust vectoring, fuel consumption, and environmental effects
• 📡 Progressive Scenarios: Easy (wide beam) → Medium (standard) → Hard (narrow beam, high noise)
• ⚡ Production Ready: FastAPI deployment + comprehensive logging for real-world applications
• 🌡️ Advanced Physics v2.0: ISA atmospheric models, Mach effects, sensor delays, thrust dynamics, and domain randomization for improved sim-to-real transfer

---

⚖️ Legal and Ethical Use Notice

This repository is released under the Hippocratic License 2.1, which permits use, modification, and distribution of this software for purposes that do not violate human rights or enable weaponized, military, or surveillance applications.

This project is for academic, research, and peaceful experimentation only.

Use of this code in autonomous weapons, missile guidance systems, surveillance infrastructure, or other military/defense-related applications is explicitly prohibited.

If you are unsure whether your intended use violates this principle, do not use this software.

---

⚖️ Legal and Ethical Disclaimer

This project is a purely academic and simulated environment intended for reinforcement learning research and experimentation. It does not interface with real-world defense systems, targeting software, or weaponized hardware.

The repository does not include any classified, restricted, or export-controlled materials. It is intended only for educational and non-military applications.

Use of this software must comply with applicable U.S. export control laws (ITAR, EAR) and international dual-use regulations. The authors explicitly prohibit any use of this project for real-world weaponization, targeting, or autonomous lethal systems.

By using this repository, you agree to use it solely for lawful, ethical, and research-related purposes.