Visual Relocalization in Sparse, Low-Texture via 3DGS (IROS 2026)
rsasaki0109 · x · 2026-08-17
A method for visual relocalization in aliased and low-texture environments via novel view synthesis. It involves offline map building where submaps are trained with 3D Gaussian Splatting using geometry-aware supervision (MVS depth/normal alignment, LiDAR Chamfer loss), and online pose estimation against the trained 3DGS models using 6DGS.
More from Embodied
- Maker completes DIY robot model with newly identified motors, ready for experiments — haydendevs · 2026-08-17
- Chinese AI laser mosquito-killing device opens pre-orders at ~$1,000 — AIFlow_ML · 2026-08-17
- Robotics blogger admits wheeled bimanual robots can't put on a show like humanoids — cixliv · 2026-08-17
- Xynova showcases Flex 2 hybrid-driven dexterous hand with 22 DoF — CyberRobooo · 2026-08-17
- Recommended workflows for AI-generated 3D models from reference images for printing? — uselessfellw · 2026-08-17
- Contactless Blood Pressure Monitoring Breakthrough: FMCW Radar with Momentum-Driven Calibration Achieves High Precision — maier_ak · 2026-08-17