Andrew Davison: robots need object-based SLAM, not scan-then-fit reconstructions
AjdDavison · x · 2026-10-07
Imperial College's Andrew Davison argues explicit scene reconstruction is only useful for robotics if fast and incremental — i.e., object-based SLAM. Instead of scanning a scene into a dense point cloud and fitting objects afterward, robots should fit objects directly and use the object graph as the scene's "map." He cites prior work ReorientBot/MoreFusion (ICRA 2022), which combined explicit reconstruction and motion planning for accurate manipulation, open-sourced on GitHub.
More from Embodied
- RaiSim is back: v2.8.0 brings ultra-fast accurate CPU simulation plus photorealistic rendering — ChongZzZhang · 2026-10-07
- LiteReality-Agent turns images into simulation-ready 3D scenes for MuJoCo robot planning — elliottszwu · 2026-10-07
- French TV host swings at two robots live on W9 in hilarious human-vs-robot bout — iamfakhrealam · 2026-10-07
- Seeed engineer to give ROSCon UK 2026 deep dive on ROS2Control interface for STS3215 servos — kamathsblog · 2026-10-07
- Menlo Research open-sources Asimov 1 humanoid's locomotion policy and Isaac Lab training code — freelerobot · 2026-10-07
- Self-Compensating VLA Boosts Robot Arm Task Success by Over 30 Points — POSTECH · 2026-10-07