LBDU-VIO Cuts Position Error by 25.1% on EuRoC During 10s Visual Outages via Learned Bias Dynamics
zhenjun_zhao · x · 2026-10-01
A new arXiv paper, LBDU-VIO, tackles visual-inertial odometry for aerial robots when vision is unreliable, upgrading conventional MSCKF filters:
- A neural ODE models continuous-time bias dynamics, replacing the random-walk assumption
- An IMU uncertainty model predicts motion-adaptive measurement noise covariances for covariance propagation
- Both models train from pose supervision alone, without direct labels
- On real-world EuRoC and TUM-VI benchmarks it beats representative baselines, cutting mean relative position error by 25.1% vs. S-MSCKF on EuRoC sequences with 10-second visual outages
More from Embodied
- Custom Retro Hardware Is a Whole New World with AI for Hardware Design — pvncher · 2026-10-01
- Researcher dismantles Figure's IP excuse for destroying F.02: "awful marketing" — MarwaEldiwiny · 2026-10-01
- Autonomous's $20,900 Solar WorkPod Offers a Backyard Office with Dual RTX 5090 AI Datacenter — dee_hw · 2026-10-01
- Boston Dynamics' four-finger hand praised: balancing capability and simplicity — MarwaEldiwiny · 2026-10-01
- EPFL's TERRA pipeline teaches muscle-actuated agents to walk 9.4 hours of diverse terrain — amathislab · 2026-10-01
- Humanoid robot does real factory work in uncut 2-hour video, now running 4-6 hour shifts — chris_j_paxton · 2026-10-01