Microsoft launches Project AirSim, the first drone simulator for training AI systems

Unlike Microsoft Flight Simulator, Project AirSim functions as a virtual world for AI systems, where software developed to fly autonomous drones can train in thousands of simulated flights.

Companies looking to use autonomous drones will now be able to train their software long before sending them to the location, anticipating any obstacles and potentially risky situations. For example, in the case of drones used for deliveries, their software will be able to be trained to avoid high-voltage cables and anticipate draughts around tall buildings.

According to Microsoft, the accuracy and adaptability of autonomous flight systems can be improved on the fast-forward, simulating millions of flights per second in which all sorts of adverse situations are analyzed. By recreating the flight device in a virtual environment, developers can learn how the vehicle will behave in the real world. For example, by simulating the effect of rain or strong winds on the stability of the aircraft, or the range offered by the battery.

Eventually, Microsoft could also use the AirSim platform to train Air Taxi vehicles, simulating flight routes before giving the green light to passenger rides or parcel delivery.

The AirSim project runs on Microsoft’s Azure cloud computing platform and already has a first customer, with US company Airtonomy turning to Microsoft’s services to train drones used to inspect wind turbines and high-voltage grid infrastructure. For example, while checking a single wind turbine used to require a crew of three people working most of the day to scale the 80-metre-high equipment, the same task can now be performed by a single employee with minimal training. Basically, all they have to do is start up the pre-trained drone in the simulator and wait for the routines included in the flight programme to run.

Although, Microsoft is talking exclusively about use in civilian applications, it is expected that AirSim (or a similar platform) will quickly find its use in military purposes as well, training drones for complex missions behind enemy lines.