News hardware Metavers: Microsoft trains drones to fly in a virtual environment
AirSim is the name of the new program developed by Microsoft: a virtual environment to train drones and autonomous aircraft to fly in real situations. A sophisticated metaverse, powered by the power of artificial intelligence.
The return of Microsoft’s AirSim program
For the most knowledgeable among you, the name “AirSim” may be familiar. It was already the name of a drone simulator, made available to developers by Microsoft Research a few years ago.. But the program, intended to train autonomous vehicles of all kinds, had not proved conclusive, and the Redmond firm had closed it prematurely.
Today in a detailed blog post, Microsoft announces the new AirSim project. The goal is the same: train drones, aircraft and other autonomous flying vehicles to move in real conditions… Via a virtual environment. The latter uses artificial intelligence to generate virtual environments that are faithful to reality, thanks to the use of a titanic amount of data, provided by the Azure cloud.
What makes this new AirSim project more attractive than its late predecessor is its accessibility: no need to be an expert in machine learning to get started, developers can now rely on a large library ofassets 3D, and on easy-to-use artificial intelligence models, to Quickly and efficiently create ready-to-use virtual flight environments.
An ultra-realistic virtual training environment
In addition to the data provided by Azure, Microsoft has worked with industry expertssuch as the company Airtonomy, to ensure that this simulator is as consistent as possible with a real flight experience.
The AirSim therefore makes it possible to train the artificial intelligence of drones, by placing them in these ultra-realistic 3D simulations, where they learn to evolve virtually in hostile natural environments such as in the skies of major world metropolises.
The drones are thus trained to react in real time: how to detect, avoid and circumvent such and such an obstacle; how to fly in rain, frost or snow; what are the consequences of wind or heat on the visibility, autonomy and precision of the drone? From simulation to simulation, AI models memorize their errors, and are able to adapt to less and less clement flight conditions.
“Gathering this data is impossible to do in the real world, where you can’t afford to make millions of mistakes. In fact, you can rarely afford to make a single mistake. — Matt Holvey, director of intelligent systems at Bell, one of the companies that participated in the AirSim program in early access.
The objective: to develop new forms of mobility
Microsoft is considering different uses for its AirSim program, which is not yet officially deployed, but which is gradually opening up to the public. The simulator can be used for the training of personal drones as well as for the training of industrial autonomous vehicles, like flying taxis or parcel carrier drones.
“Autonomous systems will transform many industries and enable many aerial scenarios, from end-of-line goods delivery in the densest cities, to inspecting downed power lines miles away. — Gurdeep Pall, vice president of Business Incubations in Technology & Research at Microsoft.
Especially since the more Microsoft opens its program, the more data it compiles, thus making it more and more efficient. Eventually, if it became open-source, the AirSim program could constitute a real reference in terms of machine learning for flying vehicles. The Microsoft article also states that NASA has already successfully tested the simulator as part of one of its projects.
“A ton of data is generated when a flying machine navigates in Project AirSim. Our ability to harvest (and transform) this data will significantly change the aviation landscape. And thanks to that, we will see more and more vehicles in the air, helping to monitor farms, inspect major infrastructure, and transport goods and people to the most remote places. — Ashish Kapoor, creator of the first AirSim project, now in charge of the Autonomous Flying Devices Research Group at Microsoft.