Last month, MAK Technologies released MAK Legion 1.0 — the first commercial version of our new Scalability and Interoperability Framework. I wanted to update everyone on some of the ways we're already using MAK Legion to achieve distributed simulations at unprecedented scale.
Many of our early Legion demos, as part of the US Army STE CSE program, were designed to show how Legion enabled MAK's own simulation engine, VR-Forces, to deliver extreme entity counts. We often connected 3rd party visual- or player-clients to our large-scale Legion exercises (including VBS4, OneSAF, the Unreal Engine, and Exonaut) to demonstrate interoperability; but VR-Forces Sim Engines typically did the heavy lifting of simulating the millions of entities that the various player clients could view and interact with.
This year, we've expanded our efforts to show that MAK Legion can bring greater scalability, as well as cloud-deployability, to 3rd-party applications on the "entity generation" side as well - including game engines, open-source simulation tools, live data feeds, and even COTS products from MAK's competitors.
For this demonstration, MAK's engineers deployed 36 copies of the standard, commercial, Windows-based VBS4 product across a total of six virtual machines running on the AWS cloud. Each instance of VBS4 was configured to run in "headless" mode, so that it could run on standard VMs without requiring GPUs; and each simulated 1,000 pattern-of-life entities. All 36 copies loaded a custom "Legion Adapter" plug-in, and communicated through a single MAK Legion Server (also running on the AWS cloud) rather than through VBS4's built-in networking protocol.
We ran one additional copy of VBS4 as a player client on a gaming laptop in our lab in Orlando. The player client, which could be configured as a UAV sensor, soldier, or ground vehicle, connected to the cloud-hosted Legion Server, and used Legion's interest management to stream in and continuously receive state updates for the closest 4,000-6,000 entities, as the player's field of view moved through the scene. The 36,000 full-fidelity VBS pattern-of-life entities were all moving, path-planning and navigating on 3D terrain, and avoiding collisions with both terrain obstacles and each other. We configured each instance of VBS4 to subscribe to a region that overlapped at least one neighboring instance's entities - to prove that our Legion world wasn't "sharded", and that the Legion Server could serve different sets of thousands of entities to all of the VBS4 instances simultaneously. We are confident that with a more direct integration and further collaboration, MAK Legion could enable VBS4 to achieve entity counts well into the millions.
MAK also recently prototyped a Legion-based application that leverages the power of the NVIDIA FleX engine to simulate up to 65,000 entities on a single laptop using compute shaders running on the GPU. The FleX engine implements collision avoidance, goal-seeking, and fleeing from threats or detonations by modeling human characters as particles that are attracted or repelled by a variety of real-time forces. The entities were published through Legion, and visualized as fully animated 3D human characters in both an Unreal-based Legion client, and in MAK's own VR-Vantage 3D engine.
Finally, MAK has shown that Legion can be used to bring live or recorded real-world data feeds into a large-scale simulation environment: we built a lightweight application called Legion-Air that loads recorded ADS-B tracks for all of the world's aircraft on a given day, and publishes those tracks as smoothly-simulated Legion entities. We used the Legion monitor to display the positions of all 30,000 - 40,000 continuously updating aircraft positions and orientations on a map of the world.