Maxence Boels

Artificial Intelligence Researcher

DragonFly DE-1 Laser Defense Turret

#ROBOTICS #DEFENSE #AI

Project Overview

The DragonFly DE-1 is an autonomous laser defense turret system combining computer vision, multi-agent reinforcement learning, and precision targeting. The project features both PyBullet simulation and real-world hardware deployment capabilities.

The system uses coordinated turrets to detect, track, and engage aerial targets with 405nm UV lasers, trained through adversarial learning algorithms for optimal performance against evasive targets.

Project Progress

3D Printed Turret Parts

Hardware assembly in progress - electronics integration pending

CAD Model of Turret

CAD model of the turret mechanism with pan & tilt capabilities

Technical Details

Hardware Components

  • 3D printed pan & tilt turret mechanism
  • High-torque digital servos with metal gears
  • OpenMV camera module with specialized firmware
  • 405nm UV laser diode (5mW)
  • Servo control PCB and drivers (pending integration)
  • Arduino-compatible microcontroller

Performance Specifications

  • Pan range: 360° continuous rotation
  • Tilt range: -30° to +90°
  • Targeting precision: ±0.5° at 10m
  • Detection range: Up to 50m for standard targets
  • Target acquisition time: <500ms
  • Classification accuracy: >95%

Software Stack

  • PyBullet physics simulation environment
  • Multi-agent PPO reinforcement learning
  • Computer vision with OpenCV
  • Real-time target tracking and prediction
  • Adversarial training for evasion countermeasures
  • Hardware abstraction layer for real-world deployment

Key Features

Multi-Agent Coordination

Multiple turrets coordinate using centralized critic with decentralized execution, enabling efficient coverage and target prioritization.

Adversarial Training

The system implements adversarial learning where turrets train against increasingly sophisticated evasive targets, improving real-world performance.

Simulation-to-Reality Transfer

Comprehensive simulation environment with physics-accurate modeling enables reliable transfer of trained policies to physical hardware.