SplatLab: an autonomous research system that writes submission-ready 3DGS papers beating SOTA by up to +3.78 dB
zhenjun_zhao · x · 2026-10-24
Researchers unveiled SplatLab, an autonomous research system for 3D Gaussian Splatting that, given a prompt, a target paper URL, or no input at all, reproduces a baseline, proposes extensions, trains and ablates across Mip-NeRF 360 and LLFF, writes the manuscript, and binds every reported number to a logged training run.
- Three-level agent hierarchy (Professor → PhD coordinators → undergrad leaf agents) runs six stages: grounding, Elo-pre-ranked ideation, EVOLVE-BLOCK plugin implementation, cascade-weighted Elo tournament across eight islands, integrity audit, and reviewer-driven manuscript synthesis
- Distinctive mechanisms: verbal-gradient feedback in GPU-measured PSNR tournaments; a standing falsifier that subtracts Elo on failure; runs cannot terminate while any claim is unbacked
- Results: generated manuscripts beat reproduced targets by +0.18 to +3.78 dB PSNR, with methods like Nebula3DGS-v2 and BubbleGS surpassing published SOTA in their regimes
More from coding & agent
- DeepMind researcher demos a playable 3D racing game built right on her desk — thursdai_pod · 2026-10-24
- Cut LLM costs: route requests, disable thinking, cache answers, escalate only on low confidence — colinmcnamara · 2026-10-24
- NVIDIA's OpenShell: Apache-2.0 runtime sandbox for autonomous agent fleets, 18.6k stars — petrusenko_max · 2026-10-24
- Claude Code Routines runs overnight, screens 380 clinics into 30 outreach drafts — Arindam_1729 · 2026-10-24
- Factory and Anthropic share customer story on long-horizon coding agents and 'dark factories' — matanSF · 2026-10-24
- Dev argues Rust rewrite of Apache2 with Verus spec beats running C code in expectation — jd_pressman · 2026-10-24