Micro Iron Dome
ARYA Integration
Digital Twin Architecture for Event-Based Counter-Drone Interception — Unreal Engine to ARYA Deterministic Physics Engine
Executive Summary
A comprehensive system architecture bridging Unreal Engine with ARYA for defense-grade drone interception simulation.
The Micro Iron Dome project, developed by three 17-year-old students from NUS High School of Mathematics and Science, has demonstrated exceptional potential in high-speed drone interception using event cameras (Dynamic Vision Sensors). Their system achieves a remarkable 3.8 ms computational latency, outperforming traditional frame-based tracking by 93-95%.
This report proposes a comprehensive system architecture that bridges their UE environment with ARYA, the deterministic multi-physics simulation engine. The goal is to create a Digital Twin of the complete interception scenario that enables rigorous physics validation, AI-driven optimization, and defense-grade safety governance.
ARYA's deterministic engine does not estimate or predict physics — it solves them using first-principles solvers with constraint-validated output. Crucially, this architecture introduces the Safety Kernel and autonomy level governance (A1-A6). For a defense system, the decision architecture ("who authorizes the intercept") is as important as the physics.
The Problem: Why UE Alone Is Not Enough
Unreal Engine excels at visual simulation, but defense-grade interception demands more.
Physics Fidelity
UE's Chaos engine uses game-grade approximations. Aerodynamic drag, Magnus effect, and wind shear are simplified or absent — insufficient for computing 200+ m/s projectile intercepts.
Sensor Modeling
UE cannot model event camera physics — photocurrent generation, threshold mismatch, dark current noise, or the luminance-to-event relationship. Algorithms trained on inaccurate synthetic data won't transfer to real hardware.
Determinism
UE's physics is not fully deterministic across hardware configurations and frame rates. Defense systems require identical results every run with the same inputs. ARYA guarantees this.
Multi-Physics Coupling
Interception involves simultaneous fluid dynamics, optics, electronics, and structural mechanics. UE cannot model these coupled domains. ARYA's nano-model architecture handles exactly this.
System Architecture Overview
A closed-loop Sim-to-Real pipeline consisting of three primary layers, connected by a robust data bridge.
Fig. 1 — Three-layer system architecture: UE5 → Data Bridge → ARYA
Unreal Engine 5 — Rendering & Synthetic Data
Data Bridge — Communication Layer
ARYA — Deterministic Physics & Safety
The CNN Boundary & Safety Architecture
The most critical architectural shift: placing the interception algorithm inside ARYA's Constraint Layer.
The CNN Cannot Directly Actuate the Interceptor
Currently, the CNN acts as an independent agent that directly commands the actuator. In a defense context, this is a dangerous architecture. Instead, the CNN must be reimplemented as a nano-model inside ARYA. All outputs must pass through the Constraint Layer, which validates proposed trajectories against inviolable physical laws before passing them to the deterministic NM-Intercept solver.
Fig. 2 — CNN Boundary: WRONG (dangerous, direct control) vs. CORRECT (inside Constraint Layer with Safety Kernel)
The Safety Kernel is the governance constant. It enforces the selected Autonomy Level and ensures no action is taken without proper authorization. As the system matures, it can graduate to higher autonomy levels — but the Safety Kernel remains.
Autonomy Level Governance (A1-A6)
ARYA Nano-Model Architecture
Each model is bound by Inviolable Constraints — physical laws the AI cannot break under any circumstances.
Computes exact projectile trajectory accounting for drag, gravity, wind shear, and Magnus effect.
First-principles kinematics & fluid dynamics
- ■Mass conservation
- ■Energy conservation
- ■Max velocity ≤ muzzle velocity + gravity assist
Models event camera sensor physics: photocurrent, threshold mismatch, dark current noise.
First-principles optics & electronics
- ■Event generation rate ≤ hardware bandwidth limit
- ■Minimum luminance threshold > 0
Simulates target drone flight dynamics (lift, drag, thrust vectors).
Constraint-validated aerodynamic solver (ROMs based on full CFD)
- ■Acceleration ≤ max motor thrust / mass
- ■Turn rate ≤ structural limit of drone frame
Models pan-tilt mechanism, servo response, backlash, and vibration.
First-principles structural dynamics
- ■Slew rate ≤ max servo RPM
- ■Range of motion within physical hard stops
Computes 3D interception geometry and optimal firing time/lead angle.
Deterministic intersection solver
- ■Projectile and target must occupy identical 3D volume at identical timestamp t
The students' sparse CNN tracker, ported into ARYA.
Neural Network (CNN)
- ■Output coordinates must fall within the physical FOV of the sensor
NM-Swarm is an architectural capability enabled by ARYA's foundation model pattern for multi-agent coordination. No swarm engagement scenario has been executed to date for this specific project, but the architecture supports its future implementation.
The Sim-to-Real Training Pipeline
A powerful closed-loop pipeline for training and validating the interception algorithms.
Fig. 3 — 9-step Sim-to-Real Pipeline with Safety Kernel at Step 7
Scenario Generation
UE5Generates a training scenario with randomized parameters (drone type, speed, weather).
Target Production Requirement: W3C PROV-DM Lineage
Every hit/miss evaluation must be logged with W3C PROV-DM provenance — an immutable audit trail of exactly which sensor data, model version, and constraints led to each engagement decision. This makes simulation results defensible to military customers.
Note: Full PROV-DM implementation is scoped out of the initial 3-day hackathon.
CAD Integration Inputs
To create a complete hardware Digital Twin, the team must provide specific inputs for ARYA's zero-shot deployment.
CAD Models
Complete assembly in Fusion360, STEP, or CATIA format (pan-tilt mechanism, barrel, camera mount).
Fusion360 / STEP / CATIAMaterial Properties
Exact materials for all structural components (e.g., Aluminum 6061, specific 3D printed polymers) to calculate mass, inertia, and thermal expansion.
Material datasheetsOperating Parameters
Servo motor torque specs, gear ratios, camera resolution, and processing unit clock speeds.
Technical specificationsMission Context
The operational envelope: max target speed 75 m/s, max range 100m, expected weather conditions.
Operational requirementsHackathon Implementation Strategy
A structured 3-day curriculum for the NUS High team to teach at the defense tech hackathon.
UE Fundamentals
Set up drone flight in UE5 and extract synthetic DVS data.
The Middleware Bridge
Implement the UDP-to-REST middleware to send UE data to ARYA without blocking the game thread.
Physics & Safety Integration
Introduce ARYA's QO interface. Port the CNN into ARYA. Demonstrate the Safety Kernel blocking an unauthorized firing command.
Crucial lesson: Demonstrate the Safety Kernel blocking an unauthorized firing command.
Full Pipeline
Run batch simulations and use the AI tuning layer to calibrate against real-world test data.
Batch Mode Fallback
If the real-time UDP bridge proves unstable with 30 participants on a local network, the hackathon will immediately fall back to Batch Mode. Participants will record UE trajectories and event streams to disk, and ARYA will process and validate them offline. This ensures the learning objectives are met regardless of network stability.
Conclusion
By integrating their Unreal Engine visual simulation with ARYA's deterministic multi-physics engine, the Micro Iron Dome team can transition their prototype into a robust, auditable simulation framework.
The most important lesson for the students is that defense technology is not just about fast algorithms and accurate physics — it is about governance and safety. By placing their CNN inside ARYA's Constraint Layer, enforcing inviolable physical laws, and governing engagements through the unfireable Safety Kernel at Autonomy Level A3, they are learning the true architecture of trustworthy defense AI.
References
J. Embley-Riches et al., "Unreal Robotics Lab: A High-Fidelity Robotics Simulator with Advanced Physics and Rendering," arXiv:2504.14135, 2025.
A. Loquercio et al., "Deep Drone Racing: from Simulation to Reality with Domain Randomization," IEEE Transactions on Robotics, 2019.
H. Rebecq et al., "ESIM: an Open Event Camera Simulator," Conference on Robot Learning (CoRL), 2018.
Duality AI, "Pioneering a Digital-First Approach, U.S. Army Contracts Digital Twin Simulation Company, Duality AI, for Development of AI-Based Anti-Drone System," April 2025.