Kapaxha - Drone Swarm Control System
Core software architecture for coordinating a swarm of drones in real-time - near-zero latency comms and YOLO-based leader drone tracking.
Project Snapshot
Role
System Design Engineer (Intern)
Organization
Kapaxha Dynamics
Timeline
October – December 2024
Status
Completed
Deployment
Internal demonstration / development environment
Tech Stack
Impact & Results
- Near-zero latency drone swarm communication achieved
- Real-time leader drone tracking with smooth Kalman-filtered trajectory
The core software architecture for coordinating a swarm of drones in real-time, designed and implemented for defence applications at Kapaxha Dynamics. It features near-zero latency communications and YOLO-based leader drone tracking.
The Problem
Coordinating a swarm of drones in real-time requires ultra-low latency communication and robust tracking to maintain formation and execute commands seamlessly in a defence environment.
The Solution
A custom communication protocol and an advanced visual tracking system that enables drones to communicate with near-zero latency and smoothly track a leader drone's trajectory.
Architecture & Implementation
Communication: Built a custom UDP/TCP protocol coupled with a dedicated control algorithm for near-zero latency drone-to-drone communication.
Leader Tracking: Implemented a YOLO-based real-time leader drone detection system.
Trajectory Smoothing: Enhanced the detection system with a Kalman filter for smooth trajectory tracking and prediction across frames.
Data Used: Real-time video feeds and telemetry data.
Tech Stack: Python, YOLO, OpenCV, Kalman Filter, UDP/TCP, TensorFlow.
Challenges & Solutions
Latency: Achieving near-zero latency communication over unreliable and noisy wireless networks.
Computer Vision: Dealing with motion blur, varied lighting, and rapid movement in drone video feeds for the YOLO tracker.
Control Systems: Tuning the Kalman filter to prevent erratic trajectory predictions.
Key Takeaways
UDP/TCP socket programming for real-time systems.
YOLO model optimization for edge deployment and fast inference.
Applied control theory (Kalman filtering) for trajectory smoothing.