severe drift (running with unitree lidar L2)

[LiqunXu]

When I run with unitree lidar L2 (I use unilidar sdk2), there is severe drift, even if I don't move the lidar, the aft_mapped frame keeps rotating and translating around the camera_init frame, does anyone have the same problem?
here is the video shows the problem: https://uwprod-my.sharepoint.com/:v:/g/personal/lxu322_wisc_edu/EdCC8NIons9Lp0nz59942Q8BjTsYPn0cR4CaP8e7kgakVw?nav=eyJyZWZlcnJhbEluZm8iOnsicmVmZXJyYWxBcHAiOiJPbmVEcml2ZUZvckJ1c2luZXNzIiwicmVmZXJyYWxBcHBQbGF0Zm9ybSI6IldlYiIsInJlZmVycmFsTW9kZSI6InZpZXciLCJyZWZlcnJhbFZpZXciOiJNeUZpbGVzTGlua0NvcHkifX0&e=o4oFU2

[dfloreaa]

That's a common problem with the PoinyLIO on the Unilidar 3D LiDARs; unless a lot of static, close-by points are found during startup of the algorithm, the odometry will start to drift and rotate around due to noisy measurement for close-by points.

I recommend having some sort of reference close to your sensor when started, then you can remove it and continue with your runtime.

[LiqunXu]

That's a common problem with the PoinyLIO on the Unilidar 3D LiDARs; unless a lot of static, close-by points are found during startup of the algorithm, the odometry will start to drift and rotate around due to noisy measurement for close-by points.

I recommend having some sort of reference close to your sensor when started, then you can remove it and continue with your runtime.

Thanks for replying, I tried but it did not work. I think this is because the delay for /unilidar/cloud, I tried to modify the "time_lag_imu_to_lidar" in the lidar config file, sometimes it worked, but still not perfect.

(base) liqun@liqun-3090:/robot_project/catkin_point_lio_unilidar$ ros2 topic delay /unilidar/cloud
average delay: 0.099
min: 0.092s max: 0.108s std dev: 0.00478s window: 9
average delay: 0.100
min: 0.092s max: 0.108s std dev: 0.00411s window: 19
average delay: 0.100
min: 0.092s max: 0.108s std dev: 0.00393s window: 29
average delay: 0.099
min: 0.091s max: 0.108s std dev: 0.00398s window: 39
^C(base) liqun@liqun-3090:/robot_project/catkin_point_lio_unilidar$ ros2 topic delay /unilidar/imu
average delay: 0.000
min: 0.000s max: 0.000s std dev: 0.00003s window: 814
average delay: 0.000
min: 0.000s max: 0.000s std dev: 0.00003s window: 1634
average delay: 0.000
min: 0.000s max: 0.000s std dev: 0.00003s window: 2453

[dfloreaa]

I do think this is a recurrent bug from the algorithm, but I will definitely take a look.

Keeping you updated

[ken4647]

I have the same problem. However, I find that it will be okay once you slightly touch the rotating part of lidar when the point_lio startup(before the drift begins). @dfloreaa @LiqunXu
But I can not find out why it will be like this.

[LiqunXu]

I have the same problem. However, I find that it will be okay once you slightly touch the rotating part of lidar when the point_lio startup(before the drift begins). @dfloreaa @LiqunXu But I can not find out why it will be like this.

Thank you. I also think that when initializing the lidar, keep it as stable as possible, and also pay attention to the CPU. If its computing power is not enough, this problem will also occur.

By the way, I noticed a lot of noise for Unitree L2 lidar. I'm not sure if this is a specific issue with my lidar or a general problem. You can see that even if it's a flat surface, there's point cloud there. Can you please check if you have this problem with yours?

[ken4647]

@LiqunXu
Thanks for your reply, but I feel like you misunderstood my meaning. ** On my device, you need to make lidar "unstable" when initializing the SLAM algorithm, or it will make severe dift (endless rotating often) **. So I think it is too absurd( may be a bug?). I try to fix it but failed actually, and it seems not be the problem from my PC's performance, as none of core of my PC work fully.

The PointCloud2 topic on my device appears normal and clear, and the noticeable drift only occurring at the beginning. The sample likes below (I put the L2-lidar on my desktop).

this resp runs like below if touch it before the SLAM begin:

The wrong one:

[newkillerwhale]

解决了吗，我也一样，启动就旋转，改参数也是有很大几率旋转，特别是空旷一点点的位置，还是没找到是什么原因

[ken4647]

解决了吗，我也一样，启动就旋转，改参数也是有很大几率旋转，特别是空旷一点点的位置，还是没找到是什么原因

It's very likely an issue with the algorithm implementation. When converting to laserscan and using slam_toolbox, this problem will not exist.

[newkillerwhale]

但是我现在就就是想使用pointlio来进行定位，我也找不到别的适配unilidar的slam算法了，unilidar这个问题真的对于我太难解决了，而且我自己改了fastlio也会出现问题，
我自己改的fastlio启动都没问题，但是在旋转一段时间后里程计也会开始漂移，用起来很难受

[newkillerwhale]

解决了吗，我也一样，启动就旋转，改参数也是有很大几率旋转，特别是空旷一点点的位置，还是没找到是什么原因

It's very likely an issue with the algorithm implementation. When converting to laserscan and using slam_toolbox, this problem will not exist.

请问您也是要调unilidar吗，我库里面改了个fastlio的，就是用起来旋转过一段时间后有几率开始飘，您有时间帮忙测试一下吗

[ken4647]

@newkillerwhale OK, I will try it torromow.

[newkillerwhale]

@newkillerwhale OK, I will try it torromow.

Can you give me some contact information like an email address?

[ken4647]

The FastLIO algorithm implementation in your repository seems to work well on my L2 LiDAR！ The drift appears to be significantly reduced, and the initial drift issue observed with the Point-LIO implementation in this repository does not occur.

@newkillerwhale OK, I will try it torromow.

Can you give me some contact information like an email address?

you can contact me with email(ken4647@outlook.com) or wechat(weekday4647).