Gaussian splatting standardization continues to ripple throughout the industry. Last week, the 26.03 release of OpenUSD marked yet another exciting addition for splats inside production pipelines. The core repository now includes a native Particle Field schema for representing 3D Gaussian Splats directly within USD’s architecture.
This revolves around the new UsdVolParticleField schema family, including a dedicated UsdVolParticleField3DGaussianSplat implementation. Proposed through the AOUSD Emerging Geometry Interest Group, the schema formalizes how Gaussian splat data can natively live inside USD without custom loaders or external wrappers.
Gaussian splats can now sit natively alongside meshes, materials, lights, rigs, and cameras in the same scene graph. Traditional geometry and photoreal volumetric reconstructions can coexist in a unified composition structure, inheriting the same layering, referencing, instancing, and override behaviors that have made USD the backbone of modern 3D pipelines.
Additionally, with splats encoded directly in USD, lifelike 3D scans can move more seamlessly into simulation and robotics workflows. As Physical AI pipelines increasingly rely on lifelike reconstructions to close the domain gap, the ability to represent these captures natively inside USD stages lowers friction between capture, simulation, and deployment. The same scene description that drives rendering can now feed training environments for robotics and spatial computing applications.
Hydra and imaging updates accompany the schema changes. A reference Gaussian splat renderer has been added to the imaging examples directory under hdParticleField, providing a concrete starting point for integrators. UsdImaging’s scene delegate mode is now officially deprecated in favor of scene index mode, aligning the imaging stack with Hydra 2.0’s architecture. Across Storm, Vulkan, and MaterialX integrations, performance optimizations and synchronization improvements continue to tighten the feedback loop between scene edits and final pixels.
The start of 2026 has seen rampant and collaborative advancement, centered around standardization and with the core belief that gaussian splatting is not going away. Please reference the OpenUSD repository for more information.
New to Gaussian Splatting? Start here
