Accord: open-source dense metric mapping from a single camera and IMU
BE2R · hf · 2026-11-20
ITMO University's BE2R Lab released Accord, a system that builds an accurate online dense metric map from just one camera and an IMU, with code and a new Phone-rig/GoPro dataset open-sourced.
- Key idea: VIO gives metric motion but sparse geometry, while feed-forward depth models give dense structure without metric scale. Accord conditions a depth foundation model on VIO filter poses, anchors each keyframe window to metric 3D points triangulated from those poses, then uses bidirectional dense depth agreement between co-visible images to estimate and correct per-image scale errors.
- Pipeline: per-window depth prediction with context keyframes, submaps fused in a factor graph.
- Results: a single untuned configuration achieves best or second-best mean F-score at 5cm on all five public benchmarks (EuRoC, TUM-VI, ORI, Hilti-Oxford, Oxford Spires) plus their Phone-rig dataset.
More from Embodied
- Zhejiang Sci-Tech University robotics lab teaches robots to roller-skate smoothly and brake — Scobleizer · 2026-11-20
- Demo shows controlling AI coding agents hands-free from smart glasses via open protocol — Scobleizer · 2026-11-20
- X-Humanoid's Tianyi 2.0 automates pressure gauge calibration 24/7 at Beijing and Jiangsu metrology labs — CyberRobooo · 2026-11-20
- Neuralink patient drives wheelchair using only his thoughts — CodeByPoonam · 2026-11-20
- Starlink Mini enables remote water gate control in mountainous Japan; all 6 tests successful — XFreeze · 2026-11-20
- O2LaF repurposes video diffusion transformers for SOTA self-supervised latent action learning — NationalUniversityofSingapore · 2026-11-20