Graphics has moved beyond drawing the screen — it has become how a robot understands the world.

Right after 3D Gaussian Splatting (3DGS) was introduced at SIGGRAPH 2023, differentiable rendering shed its graphics-tool clothing and migrated into robotics as an input representation. Gathering more than 22,000 GitHub stars in 2.9 years, 3DGS became not just a rendering technique but a new data infrastructure — one that can backpropagate gradients from pixels to pose parameters.

This hub looks, in one place, at how graphics techniques like 3DGS, differentiable rendering, and OpenUSD enable robot synthetic data, motion planning, and sim-to-real learning. It's a flow that starts in graphics and arrives at Physical AI. The wider context continues in the parent hub, Physical AI — this hub narrows in on its graphics-and-rendering axis.

In an age where a robot finds its path from a single photo and learns inside a cloud of Gaussians, what matters most? The quality of the synthetic data that this rendering and simulation produce. Pebblous is building simulation data infrastructure on top of exactly this flow.

Pebblous's Physical AI graphics position:

• ▸PebbloSim: a physics-based digital-twin simulator that generates training data for physical AI
• ▸Synthetic data quality diagnosis: verifying the distributional health and sim-to-real fitness of data produced by 3DGS and simulation
• ▸AADS (Agent Autonomous Data Scanning): quality gating inserted automatically into the rendering and simulation pipeline