search

hku-mars / FAST_LIO

A computationally efficient and robust LiDAR-inertial odometry (LIO) package
GNU General Public License v2.0
2.79k stars 939 forks source link

Too large IMU bias in estimated state. #282

Closed LarryDong closed 1 year ago

LarryDong commented 1 year ago

Hi,

I exported state estimated by esekf, and plotted all state variables. I found that the bias of IMU, e.g., ba drift too large that it become 3-4 m/s^2 after 300s. See the figure below:

plot-bias(original-fastlio)

The data is recorded by handheld avia lidar, and mapping_avia.launch is runned. The map and trajectory are both correct. More about the figure: 1) the x-axis is "the number of scans" but not the real time, so x-tick/10 is the real time in second; 2) ba is already multipled by G;

Then I test a static case using mid360 and mapping_mid360.launch. The lidar is placed stable without any movement, for 150s. Then the bias ba is also drifting to 0.8m/s^2:

plot-bias(original-fastlio-static)

I also viewed the raw IMU data, and I'm sure that the IMU does not have such severe drift. (ba-y is always 0.2m/s^2 because my IMU is a little tilted.)

So could anyone explain why this happens? Does the fast-lio estimate the bias wrong?

stale[bot] commented 1 year ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

LarryDong commented 1 year ago

My mistake. Sorry.

wyb17js commented 11 months ago

Hi bro,have you solve this problem?I meet the same question with you.This caused my position to be lost

LarryDong commented 11 months ago

Hi bro,have you solve this problem?I meet the same question with you.This caused my position to be lost

The drift is not so large as shown in the figure. I multiplied the output by 9.8 but it was already in m/s2 not g. So the drift is minor.
Besides, I don't think the large drift will lead to position lost. Please check your lidar data I suggest.

wyb17js commented 11 months ago

Hi bro,have you solve this problem?I meet the same question with you.This caused my position to be lost

The drift is not so large as shown in the figure. I multiplied the output by 9.8 but it was already in m/s2 not g. So the drift is minor. Besides, I don't think the large drift will lead to position lost. Please check your lidar data I suggest.

During the operation of my program, attitude estimation error will suddenly appear, which is related to the longer utilization time of radar data after detection. For example, my radar is 10HZ, and the radar information is used once for 0.1s, but with the increase of time, there will be an interval between the radar information used. For example, I use the radar data of 9.3s at present. Supposedly next time I should use the 9.4s incoming radar data, but in fact I may be using the 9.8s radar data, and this interval will be larger and larger, when the 1s are reached, the positioning is lost.

LarryDong commented 11 months ago

Hi bro,have you solve this problem?I meet the same question with you.This caused my position to be lost

The drift is not so large as shown in the figure. I multiplied the output by 9.8 but it was already in m/s2 not g. So the drift is minor. Besides, I don't think the large drift will lead to position lost. Please check your lidar data I suggest.

During the operation of my program, attitude estimation error will suddenly appear, which is related to the longer utilization time of radar data after detection. For example, my radar is 10HZ, and the radar information is used once for 0.1s, but with the increase of time, there will be an interval between the radar information used. For example, I use the radar data of 9.3s at present. Supposedly next time I should use the 9.4s incoming radar data, but in fact I may be using the 9.8s radar data, and this interval will be larger and larger, when the 1s are reached, the positioning is lost.

I think your issue is not IMU bias drift. Please check your raw lidar output, the lidar may fail to work under too high/cold temperature or other communication issue.
You may need open a new issue for this issue, not under current this issue.

wyb17js commented 11 months ago

Hi bro,have you solve this problem?I meet the same question with you.This caused my position to be lost

The drift is not so large as shown in the figure. I multiplied the output by 9.8 but it was already in m/s2 not g. So the drift is minor. Besides, I don't think the large drift will lead to position lost. Please check your lidar data I suggest.

During the operation of my program, attitude estimation error will suddenly appear, which is related to the longer utilization time of radar data after detection. For example, my radar is 10HZ, and the radar information is used once for 0.1s, but with the increase of time, there will be an interval between the radar information used. For example, I use the radar data of 9.3s at present. Supposedly next time I should use the 9.4s incoming radar data, but in fact I may be using the 9.8s radar data, and this interval will be larger and larger, when the 1s are reached, the positioning is lost.

I think your issue is not IMU bias drift. Please check your raw lidar output, the lidar may fail to work under too high/cold temperature or other communication issue. You may need open a new issue for this issue, not under current this issue.

OK，thank you!