Table of Contents
Production-Grade Fleet Management at Scale
This reference architecture demonstrates how to deploy and manage fleets of autonomous vehicles—drones, cars, trucks, AGVs—using open-source components and proven patterns. Every design decision targets real-world production requirements: reliability, observability, and graceful degradation.
The core architecture is vehicle-agnostic. NATS JetStream, digital twin patterns, and the authorization model work identically whether you’re managing drones or delivery trucks. Only the edge hardware and protocols change.
The Challenge
Fleet-scale operations require solving problems that don’t exist at small scale:
- Thousands of concurrent telemetry streams — Every vehicle reporting position, velocity, health, and sensor data
- Command and control at scale — Sending instructions to specific vehicles or groups without flooding the network
- Digital twin synchronization — Maintaining accurate state representation for every vehicle in real-time
- Offline resilience — Vehicles that continue operating when connectivity drops
- Safety guarantees — Ensuring network failures never compromise vehicle safety
Traditional approaches—direct protocol connections, centralized databases, polling architectures—collapse under these requirements.
Our Approach
The architecture combines three proven technologies:
| Layer | Technology | Role |
|---|---|---|
| Vehicle Control | Platform-specific (PX4, ECU, ROS 2) | Autonomous operation, failsafes, manual override |
| Edge Computing | Jetson + Industrial PC | Sensor processing, AI inference, local decisions |
| Fleet Messaging | NATS JetStream | Pub/sub, persistence, digital twin state |
Each layer operates independently. Network failures degrade gracefully—vehicles continue operating, edge computers continue processing, and state synchronizes when connectivity returns.
Architecture Components
Supported Platforms
The architecture supports multiple vehicle types with platform-specific hardware and protocols:
| Platform | Protocol | Control System | Use Cases |
|---|---|---|---|
| Drones | MAVLink | PX4/ArduPilot | Inspection, mapping, delivery |
| Ground Vehicles | CAN bus / J1939 | ECU / Autoware | Logistics, mining, agriculture |
| ROS 2 Robots | ROS 2 topics | Custom | Warehouse, research, AGVs |
Hardware Stack (Drones)
Standard hardware choices that balance capability, availability, and maintainability:
- Airframe: Holybro X500 V2 ARF — proven platform, excellent parts availability
- Flight Controller: Pixhawk 6X running PX4 v1.14 LTS
- Sensor Companion: Raspberry Pi 4 / CM4 for lightweight sensor integration
- AI Companion: NVIDIA Jetson (Orin/Xavier/Nano) for computer vision and inference
Drone Hardware Details → | Ground Vehicle Details →
Software Stack
Open-source software across every layer:
- PX4 v1.14 LTS — Flight control with proven stability
- Ubuntu Server 22.04 — Consistent Linux environment on all companions
- ROS 2 Humble — Robot middleware for sensor integration
- Go — Vehicle Gateway implementation
NATS Architecture
Hierarchical messaging topology designed for WAN deployment:
- Leaf nodes on each vehicle — local pub/sub, store-and-forward
- Regional hub clusters — aggregate vehicles by geographic region
- Global mirroring — cross-region replication when required
Digital Twin Design
JetStream streams and KV stores maintain fleet state:
- Subject hierarchy for routing and filtering
- State streams for telemetry rollup
- Event streams for audit trails
- Shadow stores for desired/reported state reconciliation
Subject Naming → | Stream Configuration →
Vehicle Gateway
Service running on each vehicle’s edge computer that bridges native protocols to NATS:
- Protocol translation (MAVLink, CAN bus, ROS 2)
- State downsampling and aggregation
- Event extraction from telemetry
- Command execution with policy enforcement
- Shadow state reconciliation
Safety Model
Network connectivity is never trusted for vehicle safety:
- Manual override is primary authority — Operator always has control (RC for drones, steering/e-stop for ground)
- Vehicle control system enforces failsafes — RTL/stop-in-place, geofence
- NATS is never in the control loop — Monitoring and coordination only
- Graceful degradation — Loss of AI or network triggers safe modes
Why This Architecture
Proven at Scale
Every component has been deployed in production environments:
- NATS powers Synadia’s global messaging infrastructure
- PX4 flies on thousands of commercial drones worldwide
- Jetson runs inference in autonomous vehicles and robots
Open Standards
No vendor lock-in:
- MAVLink is an open protocol with multiple implementations
- NATS is open-source with commercial support available
- PX4 runs on multiple flight controller hardware platforms
Your Infrastructure or Ours
NATS JetStream is 100% open source (Apache 2.0). Run it yourself or connect to our managed infrastructure—same protocol, same code, your choice.
- Self-hosted — Deploy on your infrastructure with our reference configs
- Managed — Connect your leaf nodes to our regional hubs
Free for small fleets. Scales with you.
Operational Reality
Designed for the real world:
- Vehicles can be serviced without specialized tools
- Software updates deploy over-the-air
- Telemetry data feeds standard observability stacks
- Fleet state exports to existing enterprise systems
Learn More
Explore each component in detail:
| Section | Description |
|---|---|
| Supported Platforms | Drones, ground vehicles, AGVs |
| Drone Platform | Airframe, flight controller, MAVLink |
| Ground Vehicles | Cars, trucks, CAN bus, J1939 |
| Software | Operating systems, middleware, applications |
| NATS Topology | Leaf nodes, hubs, WAN connectivity |
| Subject Naming | Hierarchical subject structure |
| Stream Configuration | JetStream setup for digital twins |
| Vehicle Gateway | Protocol-to-NATS bridge |
| Authorization & Grants | Decentralized security, third-party access |
| Safety Model | Failsafes and graceful degradation |
Get Started
Building an autonomous vehicle fleet? Deploying drones, trucks, or AGVs at scale? We can help with architecture, implementation, and operations.