Microsoft›Microsoft› offers AirSim

AirSim pairs game-engine rendering with software-in-the-loop and hardware-in-the-loop workflows, so autonomy code can be exercised against flight controllers such as PX4 before it runs on a physical vehicle. Its cross-platform, MIT-licensed distribution and precompiled environments reduce the setup work for research groups.

The plugin architecture allows custom Unreal environments and additional vehicle models and sensors without changes to the core, and the same API surface ships as a standalone library for a companion computer. Computer-vision mode supplies ground-truth depth, disparity and segmentation data that is impractical to collect in the field.

Microsoft Research released AirSim in 2017 as a simulation platform for aerial and ground autonomy research, and published its design that year in a Field and Service Robotics conference paper. The project expanded from drone support to autonomous cars, wheeled robots and static IoT devices, and was distributed under the MIT licence with precompiled environments for Windows and Linux.

In July 2022 Microsoft announced Project AirSim as its successor and stated that the original AirSim would receive no further updates. The Microsoft Research project page describes the open-source AirSim as archived, while the source code, documentation and community forum remain publicly accessible.

Open-source robotics simulators remain a common foundation for autonomy research because they allow training data to be generated and failures reproduced without field operations. Requirements have moved toward higher-fidelity environments and end-to-end platforms spanning perception, planning and testing.

AirSim is no longer maintained, and Microsoft has directed its simulation work to Project AirSim, which it presents as a commercial platform for aerial autonomy. Teams migrating from AirSim therefore compare successor platforms and other open-source simulators against the cost of adapting existing research code.

AirSim is an open-source simulator for drones, cars and other vehicles, distributed as a plugin for Unreal Engine with an experimental Unity release. Microsoft Research created it in 2017 as a platform for AI research into deep learning, computer vision and reinforcement learning for autonomous vehicles.

The simulator renders realistic environments and exposes remote-procedure-call APIs in C++, Python, C# and Java, supporting software-in-the-loop operation with flight controllers including PX4 and ArduPilot and hardware-in-the-loop operation with PX4. A computer-vision mode captures depth, disparity, surface-normal and object-segmentation images for perception work.