SLIM-init Uses Line Features and Vanishing Points for Structureless Monocular VIO Initialization, Accepted to IROS 2026
zhenjun_zhao · x · 2026-09-22
Researchers from KAIST propose SLIM-init, a structureless monocular visual-inertial initializer that exploits geometric constraints from tracked 2D line features without explicit 3D landmark reconstruction, accepted to IROS 2026.
Highlights:
- Line-derived vanishing points serve as translation-invariant orientation cues, providing rotation-only constraints under degenerate motions like low parallax or translation-dominant sequences;
- A line epipolar residual constrains translation, and a line-normal projection residual improves conditioning of linear alignment.
Unlike prior methods requiring restrictive excitation motions or expensive 3D structure reconstruction, SLIM-init outperforms state-of-the-art initialization on public benchmarks and custom degenerate-motion sequences. Source code is available.
More from Embodied
- MechFaber: Claude Code designs a 99-part quadruped with firmware co-simulated in Renode and MuJoCo — SpeedyBrowser45 · 2026-09-22
- Playing Football With Robots Is Harder Than I Thought — ChrisGPT · 2026-09-22
- What the ElevenLabs app could look like on the iPhone Duo — lukeharries · 2026-09-22
- Tesla audits Optimus suppliers across Yangtze Delta as robot orders land — XFreeze · 2026-09-22
- Spirit AI employs ~1,000 people to manipulate objects, manufacturing robot training data by hand — VraserX · 2026-09-22
- Human video emerges as robotics pretraining substrate: effective experience = hours × information per hour — rohanpaul_ai · 2026-09-22