Originally published at: Simulate Robotic Environments Faster with NVIDIA Isaac Sim and World Labs Marble | NVIDIA Technical Blog
Building realistic 3D environments for robotics simulation has traditionally been a labor-intensive process, often requiring weeks of manual modeling and setup. Now, with generative world models, you can go from a text prompt to a photorealistic, simulation-ready world in a fraction of time. By combining NVIDIA Isaac Sim, an open source robotics reference framework, with…
This is a great walkthrough—thanks for sharing. I wanted to add some context from a complementary angle.
Recently, I’ve been working on an end-to-end 3D Gaussian Splatting generation pipeline that targets large-scale, photorealistic environments, specifically designed to be simulation-ready rather than just visually impressive demos.
Unlike workflows that start from a pre-generated Gaussian scene , our approach allows users to generate high-fidelity, large-scene Gaussian PLY files directly from a renderer-driven pipeline, with consistency across scale, camera coverage, and lighting. The goal is to significantly reduce friction between content creation → Gaussian training → downstream simulation or rendering.
Some key advantages of this end-to-end approach:
-
Large-scale scene support: Optimized for room-level and building-level environments, not just object-centric captures
-
Renderer-driven consistency: Stable camera trajectories, depth accuracy, and lighting coherence, which improves Gaussian quality and downstream alignment
-
Simulation-oriented output: Clean Gaussian PLYs designed to pair naturally with collider meshes (GLB/USD), making them easier to integrate into platforms like Isaac Sim / OmniGibson
-
Fast iteration for developers: Dramatically lowers the cost of generating diverse, realistic environments for robotics, XR, or visualization use cases
I’ve shared more details about the technical journey and design decisions here:
Examples:
https://pre-share.d5render.com/team-api/shortLink/ACPOJ7
https://pre-share.d5render.com/team-api/shortLink/ZM6L1M
If anyone here is interested in testing this pipeline from a developer perspective—especially for robotics simulation, Gaussian-based rendering, or world generation—I’m happy to provide developer test accounts and discuss integration workflows.
Feel free to reply here or DM me directly. Looking forward to exchanging ideas with others working on Gaussian-based simulation worlds.
Very impressive! Is there any way to download the demo scene with a free tier Marble account?