Symbiotic Labs
A digital environment rendering engine for embodied intelligence — generative 3D assets, 4D avatars, and city-scale worlds, all served to agents and humans by a single real-time core. The pipeline that lets machines learn, test, and act inside the world before they meet it.
Text and reference images flow through a generative 3D pipeline — Gen-AI asset generation, LPM optimisation, 4D avatar insertion — into a streamed scene-block system and a real-time rendering core. One substrate for synthetic data, simulation, and the agents that train on it.
A live render of a 3D Gaussian splat scene flying through the engine at interactive frame rates — the same representation an embodied agent perceives, the same one a human can walk through. One asset, both audiences, no conversion in between.
A second pass through the splat pipeline — this time stress-testing geometry against orbit, fly-through, and close-quarters paths the original capture never saw. The artefacts that survive are the ones we train against next.
Long-vehicle dynamics, occlusion, and edge-case manoeuvres exercised inside a generated environment before a single real mile. The same scene blocks that train the perception stack drive the regression suite.