Microsoft’s Aerial Informatics and Robotics (AIR) group has created a more affordable platform for drone manufacturers to test new products: the AirSim.
Built on Unreal engine, AirSim is a platform for artificial intelligence (AI) research to experiment with deep learning, computer vision and reinforcement learning algorithms for autonomous vehicles. The simulator provides tools and makes use of recent advances in physics and perception computation to create accurate, real-world simulations.
The simulator is developed as an Unreal plugin that can you can drop into any existing, or new, Unreal environment. It is also open-source, cross platform (Windows or Linux), and supports flight controllers such as Pixhawk as well as supported protocols like MavLink, for realistic simulations.
Although quadrotors were the first (and only for now) vehicles implemented in AirSim, AIR says in the future it will come pre-loaded with commonly used aerial robotic models and several sensors, which you can see in the video below. Also, with AirSim being an open-source project, users can add custom robot models and new sensors to the simulator. AIR mentions it will add support for other type of vehicles soon.
AirSim can generate training data through a record button that writes pose and images for each frame. With the Log Viewer you can visualize the streams of data published by the simulator (and your drone). The Log Viewer’s code is also modifiable to meet your own needs.
A more complex way to generate training data is through AIR’s Application programming interfaces (API). These APIs collect sensor and ground truth data from the drones, as well as various images. With their help, the code used to test drones in the simulator can be transferred to an offboard computer and used on your real drone. Test procedures are more straightforward and easier as you are in full control of how, what, where and when you want to log data.
AirSim is a platform for AI research but it can also be a training platform. There are various options for these kinds of these training platforms which enable drone operators to take advantage of its support for different models and hardware.
AirSim is still in development but already available on GitHub. If you’re willing to help the team developing the software, AIR welcomes you to do so.