Troubleshooting¶
No unique topic¶
dedrunk topics BAG lists the candidates. Set input.topic and, for
extrinsics, trajectory.topic.
No still holds found¶
Start at rest for intrinsics.static.init and hold each orientation for
intrinsics.static.min_hold. intrinsics.static.threshold is relative to the
initial noise floor. Raising it tolerates vibration, lowering it rejects slow
motion.
Large intrinsics residuals, no gyroscope scale¶
The orientations were too similar or the rotations shared an axis. Record
more distinct holds, or use intrinsics.accel_model = "scale-bias" when only
six faces are reachable.
No random-walk slope¶
The recording is too short or disturbed. Record longer and keep
noise.cluster.max below a tenth of the duration, see noise.
Wrong gravity, translation not measured¶
The motion lacked tilt or fast rotation about several axes. A lower
lidar_imu.observability_threshold does not add information, see
results.
Too little overlap¶
The clock search removes lidar_imu.max_time_offset from both ends. Check the
timestamps, the time window and that both recordings cover the same motion.
Pelorus gets worse with the calibration¶
Check for filtered instead of raw IMU data, timestamps and units, then the transform direction and the sign of the clock offset, see the Pelorus calibration guide.