Microsoft's metaverse is for training autonomous drones

Project AirSim can allegedly run 'millions of flights in seconds'

Microsoft reckons its flight simulator platform could have autonomous drones trained up and in the sky faster than more traditional methods.

Project AirSim is built to train autonomous drones in real-world scenarios. Microsoft said it "can run through millions of flights in seconds" with drones "learning how to react to countless variables much like they would in the physical world."

Like Nvidia's Omniverse for building digital twins, Project AirSim offers 3D libraries of various environments and real-world locations. Places like New York City are plotted using Bing Maps and "other providers," and pre-trained models in Project AirSim can be used as starting points to build more customized digital drone pilots.

Simulations can be manually controlled by a pilot if required.

Also akin to Omniverse, Microsoft is using Project AirSim to talk up "the power of the industrial metaverse", said Microsoft Corporate VP for Business Incubations in Technology and Research, Gurdeep Pall.

"Autonomous systems will transform many industries and enable many aerial scenarios … But first we must safely train these systems in a realistic, virtualized world," Pall said. 

Microsoft said use cases "are almost limitless," and that Project AirSim can be used to train drones for inspections of infrastructure, mines, farmland, package delivery and even people delivery in flying taxis.

Airtonomy, which builds software that analyzes drone data for inspections, was an early tester of Project AirSim, using it to train drones to inspect wind turbines. Bell also tested Project AirSim, using it to train its drones to land autonomously in less-than-ideal conditions. 

AirSim is dead, long live Project AirSim

Project AirSim isn't Microsoft's first attempt at building a virtual environment for testing autonomous drones: That honor falls to AirSim, launched in 2017. 

Built in Unreal Engine and later ported to Unity, AirSim is open source and exposes its APIs so developers can interact with simulated vehicles programmatically. Like its successor, AirSim is able to simulate weather conditions and other scenarios to model real-world scenarios.

Microsoft said AirSim was a popular research tool, but one that required a high level of expertise in machine learning and coding. Project AirSim is the end-to-end platform version with added out-of-the-box features that "expand the universe of people who can start training autonomous aircraft," Microsoft said. 

With Project AirSim now in limited preview the original AirSim is being archived, Microsoft said. While still usable, AirSim won't receive any updates, effective immediately. Those interested in Project AirSim can visit its website to sign up for more info or get on the preview waitlist. ®