AirSim

AirSim is an open-source simulator from Microsoft Research for drones, cars and other autonomous vehicles. It runs as an Unreal Engine plugin (with an experimental Unity plugin) and exposes APIs for retrieving data and controlling vehicles. The project site states that the repository will be archived and that Microsoft is directing its efforts to a new product, Project AirSim.

What It Is

AirSim is a simulation platform for AI research, aimed at experimenting with deep learning, computer vision and reinforcement learning for autonomous vehicles. It can be dropped into any Unreal environment as a plugin. It supports software-in-the-loop simulation with flight controllers such as PX4 and ArduPilot, and hardware-in-the-loop with PX4.

How It Is Used

Vehicles can be driven manually with a remote control (drones) or arrow keys (cars), or controlled programmatically over RPC APIs from C++, Python, C# and Java. The same APIs are available as a separate cross-platform library that can be deployed on a companion computer, so code can be tested in simulation and later run on real vehicles.

Training data can be generated by pressing the record button, which writes pose and images for each frame, or by logging through the APIs. A Computer Vision mode removes vehicles and physics so cameras can be placed freely to collect depth, disparity, surface normals or object segmentation images. Weather effects can be set from a menu or through the APIs.

Current Status

The documentation announces that the original 2017 AirSim will be archived. Code remains accessible but no further updates are planned. Microsoft points users to Project AirSim as the successor, described as an end-to-end platform for developing and testing aerial autonomy. The site also notes that Microsoft and IAMAI collaborated on Project AirSim, released under the MIT license.

Docs and Ecosystem

The documentation covers building on Windows, Linux and macOS, Docker, Azure, sensors (LIDAR, distance sensor, infrared and event cameras), multiple vehicles, a ROS wrapper, PX4 and ArduPilot setup, and reinforcement learning tutorials.