Remote Drone Simulation Orchestrator
wwsim composes a versioned Gazebo world with a versioned vehicle profile, launches it on a remote GPU, and hands you a one-paste QGroundControl connection — no firewalls, no NAT, no wasted spend.
env × vehicle · MAVLink proxy · deterministic mode · auto-shutdown
The model
Two orthogonal axes, each versioned in its own repository. Swap the drone to see how an airframe handles a scene. Swap the world to see how an autonomy stack handles terrain. Neither copy-pastes the other.
Capabilities
Pick a Gazebo environment and a vehicle profile from independent registries — each a private GitHub repo, pinned to a git tag. A sim launched today is byte-identical six months from now.
A composed simulation provisions on a remote RTX-class GPU and is ready before you have switched windows. A warm pool makes hot launches near-instant.
No firewalls, no NAT, no raw UDP. Every sim returns a single TCP connection string that QGroundControl and Mission Planner accept as-is.
Attach your autonomy or navigation image as a second container, pre-wired to the MAVLink endpoint and sensor topics — isolated from every orchestrator credential.
Pin a seed and a fixed sensor-noise profile. Two runs of the same env, vehicle, and autonomy image produce identical traces — the signal your experiments optimize against.
Every sim ends with a bundle: telemetry log, parameter dump, exact env and vehicle manifests, optional rosbag and scorecard. Downloadable from the dashboard.
How it works
Choose an environment and a vehicle profile. The orchestrator resolves both manifests, checks the vehicle satisfies the env, and records the exact refs.
A remote GPU pod boots the runtime — SITL, Gazebo, MAVLink router — and, optionally, your companion autonomy container.
Paste the connection string into QGroundControl or Mission Planner and fly. Telemetry streams to the dashboard in real time.
On stop, collect the artifact bundle. Same env, vehicle, and seed, every time — so two algorithm runs are actually comparable.
Built for autonomy iteration
Companion compute, deterministic seeds, and artifact bundles are first-class — so an algorithm change is something you can measure, not just observe.
Guardrails
Sign in with an approved account to launch your first simulation.
Sign in to wwsim