patrickelectric/airsim

AirSim is a simulator for drones (and eventually other vehicles) built on Unreal Engine. It lets you fly drones in realistic virtual environments, either manually with a remote controller or programmatically through code. The project, originally from Microsoft AI & Research, is designed as a tool for AI research and experimentation with autonomous vehicles. At a high level, AirSim works as a plugin that drops into any Unreal Engine environment. You can plug in a real drone flight controller (like a Pixhawk) via USB and fly a virtual drone using your normal RC controller. The simulator provides visually realistic environments and physics. It also exposes APIs that let your code retrieve sensor data, images, and ground truth information from the drone, as well as control the vehicle. A key feature is that these APIs are platform-independent, so code you write and test in the simulator can later run on real drones without modification. This tool is built for researchers and engineers working on autonomous vehicles, particularly those using deep learning, computer vision, or reinforcement learning. For example, if you're training an AI to recognize objects while flying or navigate autonomously, you can use AirSim to gather training data (images and sensor readings) by pressing a record button or writing code to capture exactly what you need. You could also prototype and test autonomous flight logic in simulation before deploying to real hardware. A notable design choice is the focus on transfer learning, write and test code in simulation, then execute it on real drones. The API library is cross-platform and can be deployed on an onboard computer, smoothing the path from simulation to real-world deployment. The project is in beta with APIs subject to change, and the team actively welcomes contributions.