Drone Garage is a decentralized UAV communication and coordination system designed for resilient multi-drone operations in environments where traditional centralized communication may fail or become unreliable. The project combines wireless mesh networking, autonomous leader election, MAVLink telemetry forwarding, and edge-based object detection to create a distributed aerial communication platform with no single point of failure.
Modern UAV systems often rely on centralized infrastructure or direct point-to-point communication. This project explores an alternative architecture where drones communicate through a self-healing mesh network capable of maintaining connectivity even when nodes fail or move out of range.
Each drone acts as both a communication endpoint and a routing node within the network. The system dynamically elects a leader drone responsible for forwarding telemetry and detection data back to the ground station.
The platform also integrates edge AI object detection using YOLOv8 running directly on Raspberry Pi hardware.
- Decentralized mesh-based drone communication
- Dynamic leader election with automatic failover
- MAVLink telemetry forwarding across the mesh network
- Edge AI object detection using YOLOv8
- Wireless multi-hop communication
- Base station telemetry aggregation
- Low-cost hardware architecture
- Real-time detection forwarding
- UART communication between microcontrollers and onboard computers
- ZMQ-based data transport to the ground station
refer to the README in pi_code/
The system consists of several major components.
Each drone contains:
- Runs object detection
- Handles MAVLink routing
- Communicates with ESP8266
- Maintains mesh network participation
- Relays packets between drones
- Participates in leader election
The wireless mesh network is built using the painlessMesh library on ESP8266 devices.
Capabilities include:
- Multi-hop packet forwarding
- Dynamic topology management
- Node discovery
- Leader reelection after node failure
- Redundant communication paths
The elected leader drone is responsible for:
- Aggregating drone telemetry
- Receiving object detection data
- Forwarding data to the base station
- Acting as the gateway between the swarm and operator
If the leader fails, a new leader is automatically elected.
The ground station receives data through a ZMQ socket and processes:
- MAVLink telemetry
- Object detection results
- Mesh communication statistics
- Drone status information
The system uses YOLOv8 for onboard object detection.
Detection data includes:
- Object label
- Confidence score
- Bounding box coordinates
- Timestamp information
Detections are generated locally on the Raspberry Pi and transmitted through the mesh network to the base station.
- Drone captures image/video frame
- YOLOv8 performs object detection locally
- Detection packet is generated
- Raspberry Pi sends packet to ESP8266 over UART
- ESP8266 forwards packet through mesh network
- Leader drone receives packet
- Leader forwards packet to base station
- Ground station processes and displays results
Communication between the Raspberry Pi and ESP8266 uses a custom UART framing protocol.
| Field | Description |
|---|---|
| Sync Bytes | Packet synchronization |
| Length | Payload size |
| Sequence Number | Packet ordering |
| Payload | Telemetry or detection data |
| Checksum | Error detection |
- Raspberry Pi
- ESP8266
- UAV platforms
- WiFi mesh networking hardware
- Python
- C++
- YOLOv8
- MAVLink Router
- painlessMesh
- ZeroMQ (ZMQ)
The system was tested under multiple networking conditions including:
- Line-of-sight communication
- Non-line-of-sight communication
- Multi-hop relay scenarios
- Leader failure and reelection
- Variable node spacing
Metrics evaluated included:
- Packet loss
- Throughput
- End-to-end latency
- Mesh reliability
This project investigates:
- Resilient UAV swarm communication
- Distributed aerial networking
- Low-cost autonomous coordination
- Edge AI integration for UAV systems
- Fault-tolerant drone communication architectures
Potential future improvements include:
- Scaling beyond three nodes
- Autonomous flight integration
- Improved routing optimization
- Enhanced security mechanisms
- Long-range radio integration
- Distributed task coordination
- Edge model optimization
- Real-time mapping and tracking