Pow3R-SLAM runs 1.6x faster than MASt3R-SLAM with 15% lower trajectory error
zhenjun_zhao · x · 2026-10-01
Pow3R-SLAM is a real-time RGB-D SLAM system that uses Pow3R for tracking and mapping, feeding depth as a prior on the network's two-view reconstruction predictions rather than fusing it geometrically, addressing sparse depth images.
- On 24 sequences from TUM, 7-Scenes and Replica under the MASt3R-SLAM protocol: 1.6x faster wall time, 15% lower mean trajectory error, 3.1x lower unscaled error, and 30% lower Chamfer distance with denser maps.
- A hybrid variant runs at 25.3 FPS, 2.1x faster than MASt3R-SLAM while improving accuracy.
- Beats ORB-SLAM3 (RGB-D) on accuracy and completes every TUM sequence. Code to be open-sourced upon acceptance.
More from Embodied
- Dollhouse Robots hiring Field Operations Manager in SF Bay Area for robot deployment — carlosdponx · 2026-10-01
- FINGR: a dexterous hand solves a 2×2 Rubik's Cube with continuous finger tricks — HaozhiQ · 2026-10-01
- Dyna Robotics' Taku adapts to an unseen server room with just a few demos — JasonMa2020 · 2026-10-01
- Roboticist questions Figure's lava robot tests as unrealistic for home safety — MarwaEldiwiny · 2026-10-01
- Most pure robotics deployment firms aren't profitable, says founder who's talked to many — ATTlKA · 2026-10-01
- Pollen Robotics: 10,950 Microducks to ship between December 10 and January 10 — andimarafioti · 2026-10-01