Roadmap
What we've built and where we're going.
Completed
DroneOS SDK
2025Custom C++ drone control library built on ROS 2 and PX4 Autopilot. High-level APIs for flight control, state tracking, and multi-drone coordination. CLI and Python API for programmatic access. Docker-based deployment for simulation and real hardware.
Real Hardware Deployment
2025Deployed on custom X500 airframe with Pixhawk flight controller and Raspberry Pi 5 companion computer. Cloud-to-drone communication over 4G via Tailscale VPN. Production Docker containers on the companion computer.
Autonomous AI Dispatch (Simulation)
2026Integrated OpenClaw AI agent for autonomous fleet command in PX4 SITL. AI receives emergency incidents, evaluates live fleet status, selects the optimal drone, and flies it — no human pilot. Incident to airborne in ~10 seconds. Multi-drone coordination with real-time web command center.
Next — Simulation
Object Detection in Simulation
Connect the object detection pipeline to simulated camera feeds. Drone sees something in Gazebo, detection runs, agent reacts. Infrastructure exists — needs integration with the dispatch workflow.
Multi-Drone Coordination
Move beyond basic multi-drone dispatch to obstacle avoidance and airspace deconfliction. Requires virtual sensors and coordination logic. Comes after single-drone object detection is working.
Next — Framework
Modular Plugin Architecture
Restructure DroneOS into swappable modules — flight stack adapters, perception backends, AI integrations. Today it's a well-structured SDK; the goal is a true plugin system where components can be added or replaced independently.
Rust Rewrite
Rewrite the core SDK in Rust for safety, performance, and modern tooling. The current C++ codebase works but Rust offers memory safety guarantees that matter for flight-critical software.
Multi-Flight-Stack Support
Currently PX4-only. Abstract the flight stack interface so DroneOS can work with ArduPilot and other autopilot systems without rewriting the control layer.
Developer Experience
Documentation, quickstart guides, and tooling to make DroneOS usable for developers outside the project. Today it requires deep ROS 2 and PX4 knowledge to get started.
Next — Hardware
4G Reliability for AI Dispatch
Basic drone control works over Tailscale VPN. The full AI dispatch loop (incident to AI decision to flight command over cellular) hasn't been tested on real hardware. Can't be solved in simulation — must be validated on the physical drone.
Object Detection on Real Hardware
Google Coral Edge TPU on the Raspberry Pi 5. Infrastructure installed, example inference runs successfully. Needs real model tuning and integration with the drone agent for live detection.
Obstacle Avoidance Sensors
Dedicated sensors for obstacle detection in real-world flight. Evaluating options including rangefinders, lidar, and real-time computer vision. Required before any autonomous flight in uncontrolled environments.