Why Autoware all-sensors-bag4 scores 83¶

Overall: 82.8/100 · Domain: Autoware · ROS 2 · bagx 0.6.0

This is not a "bad bag"¶

The Leo Drive Isuzu all-sensors-bag4 is an official Autoware open-data recording (17.8 s, 17 topics, 9,489 messages). The Autoware domain checks pass almost entirely. What pulls the overall score down is the IMU sub-score (and, less, camera↔camera sync), not missing topics.

Score breakdown¶

Overall is the mean of the sub-scores that apply to the bag:

Component	Score	Why
GNSS	99.9	`/gnss/fix` has a 100% fix rate
Domain (Autoware)	95.0	LiDAR at 9–10 Hz ✔, GNSS ✔; the three `lucid_vision` cameras run at 14 Hz, just under the 15 Hz perception target
Sync	87.4	Camera↔camera pairs show 24–32 ms mean and up to 193 ms max offset; GNSS, IMU and LiDAR pairs stay under 4 ms mean
IMU	48.8	Accel noise 0.43 m/s² (x axis 0.57), gyro y 0.048 rad/s
Overall	82.8	(99.9 + 95.0 + 87.4 + 48.8) / 4

Reading the IMU number¶

bagx estimates noise as std(diff(x)) / √2. On a moving vehicle that number includes real vibration and road motion, not only sensor noise, so it is an upper bound. The bag is also too short (<60 s) and not static, so bagx cannot run Allan variance and says so. Treat the IMU warning as "do not take noise densities from this bag", not "the IMU is broken".

Reading the sync number¶

bagx pairs each message of the sparser topic with the nearest message on the other topic. The three cameras are not hardware-triggered against each other, so 14 Hz cameras with free-running phase sit 24–32 ms apart on average. For multi-camera perception that is worth knowing; for LiDAR localization it does not matter (each concatenated pointcloud has an /sensing/imu/imu_data sample within 3.6 ms on average).

Before the unreleased sync fix, bagx walked whichever topic sorted first by name. For a dense/sparse pair that measured the sparse topic's period instead of the alignment, which put this bag's sync at 67.7 and its overall at 77.9.

Planning / control¶

bag4 is a sensing/localization bundle. bagx notes that planning/control topics are absent and skips those checks instead of penalizing them, which is why the domain score stays at 95.

When to use this bag¶

Sensor bring-up and LiDAR/GNSS localization regression tests
bagx regression on Autoware sensing detection
Not as a source of IMU noise parameters, and not as an end-to-end Autoware gate (use driving_20_kmh, 99.0, for that)

Reproduce¶

aws s3 sync s3://autoware-files/recordings/bags/2022-08-22_leo_drive_isuzu_bags/all-sensors-bag4_compressed/ ./all-sensors-bag4_compressed --no-sign-request
bagx eval all-sensors-bag4_compressed