Microsoft’s Project AirSim tests and trains AI-powered aircrafts

Listen to this article

Microsoft launched Project AirSim, a high-fidelity simulation platform for building, training and testing autonomous aircrafts, at the Farnborough International Airshow. The platform is available now in limited preview. 

Project AirSim provides a realistic environment for AI models to run through millions of flights in seconds. AI models use these simulations to learn how to respond to countless different variables that could occur in the real world, like the weather, how temperature or winds could affect battery life. 

Airtonomy is a North Dakota-based company that has been using Project AirSim’s simulations to train autonomous aerial vehicles to inspect wind farms, survey wildlife and detect leaks in oil tanks. 

“You don’t want to fly drones into wind turbines, powerlines or really anything for that matter,” Josh Riedy, CEO of Airtonomy, said. “Coupled with the fact that winter can literally last 7 months in North Dakota, we realized we needed something other than the physical world to design our solutions for customers.”

Running on Microsoft Azure, the platform generates large amounts of data to train AI models on which actions to take during flight from takeoff to cruising to landing. Project AirSim also comes equipped with libraries of various simulated 3D environments and customizable, pretrained AI building blocks to help accelerate autonomy. 

These pretrained AI building blocks include advanced models for detecting and avoiding obstacles and performing precision landings. These out-of-the-box capabilities mean that you won’t need deep learning expertise to start training an autonomous aircraft with the platform. 

“Everyone talks about AI, but very few companies are capable of building it at scale,” Balinder Malhi, engineering lead for Project AirSim, said. “We created Project AirSim with the key capabilities we believe will help democratize and accelerate aerial autonomy – namely, the ability to accurately simulate the real world, capture and process massive amounts of data and encode autonomy without the need for deep expertise in AI.”

Microsoft hopes that the platform could be used beyond training AI models. According to Gurdeep Pall, Microsoft corporate vice president for Business Incubations in Technology & Research, Project AirSim could be used to help with the certification of autonomous systems by creating scenarios that an autonomous vehicle must be able to successfully navigate. 

Along with inspecting wind farms and surveying wildlife, Project AirSim has been used by Bell to hone its drone’s ability to land autonomously. With Project AirSim, Bell could train its AI model on thousands of different landing scenarios in minutes.

“AirSim allowed us to get a true understanding of what to expect before we flew in the real world,” Matt Holvey, director of intelligent systems at Bell, said. “It’s going to be one of the tools that will accelerate the timeline for scaling aerial mobility. If we have to test and validate everything by hand, or in a physical lab, or on a flying aircraft, we’re talking about decades, and it’s going to cost billions. But Project AirSim pulls that forward through high-fidelity simulation.”

Project AirSim began in 2017 as simply AirSim, an open-source project from Microsoft Research. The project is being retired now that the open-source tool has been repurposed into the Project AirSim end-to-end platform.