// the find
microsoft/AirSim
Open source simulator for autonomous vehicles built on Unreal Engine / Unity, from Microsoft AI & Research
AirSim is a high-fidelity simulator for drones and cars built as an Unreal Engine plugin, with a Python/C#/Java API for programmatic control. It was designed for autonomous vehicle AI research — training perception models, testing flight controllers, and running reinforcement learning experiments in simulation before touching real hardware. The project is effectively dead: Microsoft announced archival and stopped commits, pointing users toward a closed commercial successor.
The physics simulation is genuinely good — sensor noise models (IMU, GPS, barometer, magnetometer) are parameterized realistically, not just added as afterthoughts. The RPC API design is clean: same client code runs against the simulator or a real vehicle companion computer, which is the right abstraction for sim-to-real transfer. PX4 and ArduPilot hardware-in-the-loop support means you can test actual flight firmware, not a toy stand-in. The Computer Vision mode (freely poseable cameras, depth/segmentation/disparity outputs) is useful for dataset generation independent of any flight task.
The repo is abandoned — no commits, no bug fixes, and Microsoft explicitly said so in the README. Adopting this for a new project means inheriting a frozen codebase against a rapidly evolving Unreal Engine. The Unity backend was always experimental and is effectively untested. Build complexity is high: Unreal Engine setup alone is a multi-hour process before you write a line of code. The open-source successor (iamaisim/ProjectAirSim) is a community fork, not the 'Microsoft Project AirSim' commercial product, so there's no clear migration path to the thing Microsoft is actually investing in.