With 12 days and counting to release of Microsoft Flight Simulator 2020, the development updates have been getting shorter. However, more news came out regarding the SDK and third party developments. With many developers getting their hands on the new tools, new features and documentation are being developed accordingly. Developer mode also received some fixes, along with some more minor changes. Also to note, the next development roadmap for the simulator will be released between August and November, containing many of Asobo’s plans post release.
Moving into the latest partnership episode, this time featuring Blackshark.ai. Blackshark.ai is responsible for the photorealistic buildings on the ground, which have become a staple of the new flight simulator. Based in Graz, Austria, Blackshark uses a “unique artificial intelligence solution to extract precise information from satellite imagery”, which is used to construct the massive amount of accurate 3D buildings in MSFS.
While interviewing Blackshark.ai CTO Thomas Richter-Trummer, the subject of coverage came up. Richter-Trummer points out that while major cities had already been created using photogrammetry, “99.8% of the planet’s buildings, vegetation, and infrastructure” were not available in 3D. This, he says, was the challenge Blackshark sought to take on.
To overcome this challenge, a large team of AI specialists, geospatial engineers, data scientists, and rendering developers were brought in. The end result took the combined form of the Azure Cloud and artificial intelligence analyzing Bing Maps data. Using deep learning, buildings, vegetation, and roads around the globe are segmented and classified. With a patented method, buildings are reconstructed to resemble their real world counterparts (have “recognition value”), and are able to be streamed in real time.
Next up from Blackshark.ai was Stefan Habenschuss, head of machine learning at the company. Stefan points out that over the past three years, the team had reconstructed approximately 1.5 billion buildings, and detected over 30 million km2of vegetation. In places where data input is limited, AI takes over and “fills in the gaps” as Stefan puts it. The result of all this work is the player feeling immersed, as if they were flying over the real world.
Using the aforementioned deep learning tools, building footprints, height, rooftops, and type, are segmented with “high accuracy”, while facade features are added based on location and context. Stefan’s explanation of this technology was followed by Blackshark’s cloud solution architect, Fabian Schlager. He says that Microsoft and Blackshark.ai are “raising the bar” with their state of the art cloud based technology.
Schlager then discusses the power at their disposal. Hundreds of virtual machines are used in parallel to process the entire planet for global feature detection and reconstruction. This process takes a mere 72 hours due to the computing power available. Several petabytes of data are ingested, and over 1.5 billion building footprints, vegetation, and infrastructure features are output. The speed of Azure allows new AI algorithms to be implemented across the entire system in less than 30 minutes. Schlager says this is just “one of the many advantages” of using Blackshark’s technology in conjunction with the Azure cloud.
As the video comes to a close, the narrator says that Blackshark.ai is looking forward to adding new technologies to Microsoft Flight Simulator in the years to come, indicating that the simulator will improve with age.
Threshold encourages informed discussion and debate - though this can only happen if all commenters remain civil when voicing their opinions.