GaME (CVPR 2026): Gaussian mapping that forgets stale geometry as robot scenes change
lucacarlone1 · x · 2026-09-21
- Researchers from the University of Amsterdam and MIT (including Luca Carlone) unveiled GaME (Gaussian Mapping for Evolving Scenes), a CVPR 2026 paper on robot mapping with 3D Gaussian Splatting that adapts to scene changes.
- Key insight: the world changes outside a robot's camera view (e.g., a chair moved in the kitchen), yet most 3DGS-SLAM systems treat old observations as reliable constraints. GaME instead lets the map forget stale geometry and updates itself when new observations contradict old state.
- Method: a keyframe manager triggers a Dynamic Scene Adaptation (DSA) module that integrates newly observed geometry, removes outdated geometry using covisible keyframes, masks stale regions, and runs local window optimization.
- On the Flat and Aria datasets, GaME outperforms SplatAM, DG-SLAM and MonoGS on reconstructing moved objects, relocated pictures and furniture.
- Project page includes arXiv paper, code and video.
More from Embodied
- ZuckOff App Detects Nearby Meta Smart Glasses via Bluetooth Fingerprints, Tops 5,000 Downloads — LexiLove · 2026-09-21
- Stanford RL method fixes VLA latency, lifting robot success from 42% to 97% with 10 minutes of data — burny_tech · 2026-09-21
- Deadlifting 160kg at LUDUS: a striking strength demo — AIFlow_ML · 2026-09-21
- DeformSmith generates physics-grounded deformable assets for robot manipulation — Can Li · 2026-09-21
- Coowa breaks down COOWAM: on-device world model behind 10,000+ deployed city robots — 创业邦 · 2026-09-21
- Tesla's Texas Optimus line nears structural completion; Qiyuan claims a robot every 2.5 minutes — 创业邦 · 2026-09-21