SPARC

Immersive training is usually built as a set of one-off scenes. Changing a procedure means a developer, a rebuild and a release, so the person who knows the procedure cannot touch the training and the training falls behind the work it describes.

The second constraint is hardware. Content built for a headset reaches the people who have one, which in most organizations is a room, not a workforce.

SPARC separates the three things that usually get welded together: a media library, a lesson stored as data, and a renderer that reconstructs it. An instructor uploads media, places interactive elements against it and publishes. A learner opens that lesson in a VR headset or in a browser with no headset at all.

AFWERX SBIR Phase II immersive training for the US Air Force, contract FA8649-23-P-0466, $1,150,000, performed February 2023 to April 2024 and awarded to Quantum Interface LLC as prime.

Frank Robles was brought onto the proposal team and the proposal won. The work then started from zero code. He specified a team of six and, for a long stretch, built the application alone, then hired a Unity developer and added a part-time designer, and served as development lead throughout.

In his own words: after winning the SBIR, he took a product idea with zero code completed to a working solution, meeting or exceeding numerous deliverables for the Air Force. The delivered capability included gaze and motion control, so an airman with both hands occupied can move through a checklist.

Robles Industries was not the awardee and does not claim the contract as its own prime past performance. We put that sentence on the sheet because a government buyer will check, and because a vendor who overstates a role once will be assumed to have done it elsewhere.

Two things are proved by this record. The first is that a federally funded prototype can be carried into a product the firm owns, which is the path a research-to-commercial buyer is trying to walk and rarely sees walked.

The second is the shape of the thing we built. We build authoring tools rather than scenes. The separation between a media library, a lesson held as data and a renderer that reconstructs it is what lets a customer's own subject matter experts keep producing training after we leave, in a headset or in a browser, without a developer in the loop.

The authoring side is a web application. An author uploads assets tagged by type: 2D video, 360 video, image, PDF or audio. Every asset carries a name, description, internal notes, creator, publish date, version, a thumbnail that can itself be another asset, and a JSON block of asset events. Uploads go straight to object storage from the browser rather than through the application server.

A lesson is an augmented document: title, description, notes, version, view counter, author, last-modified-by, thumbnail, and a canvas asset that sets the backdrop. A 360 video puts the learner inside a scene; an image or a PDF gives a flat surface to annotate.

Onto that canvas the author places augment events, stored as JSON. A lesson also carries diagnostic and feedback parameter blocks, which is where assessment lives: what the learner is measured on, and what they are told in response.

The delivery client is a Unity application with an augment factory that instantiates each placed element from lesson data, a document manager that loads a lesson by id, a raycaster and laser pointer for selection at a distance, and a video subsystem built on a commercial plugin so 360 and 2D playback behave in-scene. Panels cover assets, documents, image display, text annotation, video control and overlay, login, tools, and points of view for moving between fixed vantage points.

The browser build is driven from the page around it. JavaScript hands the Unity player its API root, login URL, endpoints, the signed-in user's credentials and token, and the user level that decides whether this session can author or only consume, then calls preload with a document id. That is what lets one lesson library serve a headset and a browser tab without a second build of the content.

A Django project with DRF, PostgreSQL, direct-to-S3 uploads, Docker for local work and an Elastic Beanstalk deployment configuration. The REST router exposes users, groups, assets and augmented documents, with a custom user model carrying training-specific role levels. The client is Unity and C#, building to WebGL today.

SPARC was built after the contract ended, on the ideas the Air Force work proved out, with help from the same Unity developer. It is a new codebase, owned outright by Robles Industries.

• Django
• Django REST Framework
• PostgreSQL
• Amazon S3
• Unity
• C#
• TextMesh Pro
• WebGL
• Docker
• Elastic Beanstalk

A working demonstration, shown at a military conference in Canada, drew interest, and a major defense prime has asked to see it. Completing the application is the next step, and we say so plainly. The backend is Django, and the Unity project as it stands builds for the browser, with the headset target needing its XR package restored before a headset demonstration.