Skip to content

Why EuRoC MH_01_easy IMU looks noisy

Overall: 60.3/100 · Domain: SLAM · ROS 1 · MAV handheld sequence

The headline number

EuRoC is a classic visual–inertial benchmark, but MH_01_easy scores only 60.3 overall because IMU quality lands at 23.1 while sync remains strong at 97.6.

What drives the low IMU sub-score

bagx estimates IMU noise via std(diff(x))/√2 on accelerometer and gyro streams. On this bag:

  • Accelerometer noise is flagged at 0.79 m/s² — far above SLAM-oriented thresholds.
  • The platform is a flying MAV with aggressive motion; diff-based noise includes real dynamics.
  • EuRoC IMU units are older MEMS hardware; many LIO papers apply extra filtering anyway.

bagx therefore recommends LiDAR-only or LiDAR-forward odometry rather than tight LIO with default IMU weights.

Why sync still scores well

LiDAR and IMU timestamps are reasonably aligned (97.6 sync score). The recording is usable; the warning is about fusion strategy and IMU preprocessing, not a broken bag.

Practical takeaway

Goal Recommendation
VIO / LIO benchmark Expect IMU tuning; compare against NTU VIRAL pattern
LiDAR odometry only Reasonable candidate — sync is clean
Default LIO noise model Not ideal without recalibration or denoising

Reproduce

pip install bagx[ros1]
export BAGX_SCOREBOARD_BAGS=/path/to/bags
bagx eval MH_01_easy.bag

Compare with TUM VI calib-imu1 (95.4 overall) for a calibration-grade reference.