Niantic Spatial launches Gaussian Splat Relighting beta for embodied AI simulation
Scobleizer · x · 2026-09-19
- Niantic Spatial introduced Gaussian Splat Relighting (beta, design partners only): a way to generate lighting, weather, and time-of-day variants on real-world Gaussian splat captures.
- Method: an aligned mesh supplies geometry to simulate new illumination, transferred onto splats via precomputed radiance transfer (building on the 2024 PRTGS paper) — only per-Gaussian spherical-harmonics coefficients are recomputed.
- Collision meshes, navigation meshes, and semantic labels carry over untouched, so real-to-sim variants keep identical geometry and semantics — aimed at training embodied AI under controlled scene variation. Not a self-service API yet.
More from Embodied
- Robotics data and evals demand massive VLM and human-in-the-loop annotation work — paigeinsf · 2026-09-19
- WetRobo: coding agents as the interface between foundation models and lab instruments — sherryyangML · 2026-09-19
- WetRobo ships robot + demos + AGENTS.md, letting a coding agent adapt tools on site — sherryyangML · 2026-09-19
- VITURE teases VONDER AI glasses, full launch set for September 28 — Scobleizer · 2026-09-19
- ROS Weekly: ROSCon Global next week in Toronto, InOrbit open-sources Robot Ops — kscottz · 2026-09-19
- MOSS robot demo replays real Jev decisions in simulation, switching tasks from cans to bottles — kamathsblog · 2026-09-19