How are the results for calibrating sensors mounted to cars

[RichardYann]

The NOTE at the beginning said, "Accurate results require highly non-planar motions, this makes the technique poorly suited for calibrating sensors mounted to cars."

Anyone used this tool on calibrating sensors on cars?

How are the results?

[dezhi-l]

• I think it is also equally applicable to the calibration of autonomous vehicle. I test it with the sensors on own self-driving car, This tool can pretty well calibrate the external parameters between lidar-imu，The lidar-imu offset accuracy of the map built by the parameters after calibration can reach less than 10cm, and the attitude accuracy can be less than 1 degree.

• The map construction accuracy through this external parameter is very high, as shown in the figure below, The vehicle at the same position has been scanned multiple times, and the poles in the point cloud are very clear，At the same time, the position of the pole and the handheld rtk（red point） fit well

1. Pole and sign after multiple scans of the vehicle

2. Mapping

[RichardYann]

@dezhi-l Well, thank you so much!

[ry4nzhu]

Hi, I am wondering what is the configuration setup for getting the lidar-imu calibration accuracy you mentioned for autonomous vehicles. Are you using the default parameters?

[dezhi-l]

Most of the parameters just use the default parameters. I have pasted my parameters below, and it should be noted:

1. the cloud rosbag needs to contain loopback data, so as to ensure that the position(x y) and attitude(y p r) of each dimension are constrained，and the calibration accuracy will be better.
2. Both global optimization and local optimization can be used for calibration，and two different calibration methods, the optimized parameters ​​are almost the same. You just need to pay attention to the following two parameters, and set them as large as possible. @ry4nzhu
   • translation_range ( Search range around the inital_guess during the local optimization stage. ps: If you are using global optimization, the inital_guess are zero )
   • local_knn_max_dist && global_knn_max_dist (Error between points is limited to this value during global optimization.)

launch file <?xml version="1.0" encoding="UTF-8"?>

$(arg inital_guess)

[ry4nzhu]

Thanks a lot for your quick response! I really appreciate it.

I got [lidar_align-2] process has died [pid 23830, exit code -9, cmd /home/ryanzhu/lidar_pose_align/devel/lib/lidar_align/lidar_align_node __name:=lidar_align __log:=/home/ryanzhu/.ros/log/a5456592-1af4-11ee-8eaf-9905337355e7/lidar_align-2.log]. for setting the parameter "keep_points_ratio" any number larger than 0.01. Have you encountered this error before?

[dezhi-l]

sorry, I haven't encountered this issues before. but i think that the node died because the ram is full. You can open system monitor to observe the memory usage @ry4nzhu

[ry4nzhu]

Thanks @dezhi-l. Just one quick follow-up question, in your previous note, you mentioned that the cloud rosbag needs to contain loopback data. Does that mean the vehicle trajectory needs to be circular (i.e. it re-enters its beginning location to form a close loop)?

[dezhi-l]

yes, you are right. the vehicle trajectory needs to be circular. @ry4nzhu I pasted the trajectory example below

[huangwefeng121]

你好，你的数据是lidar和纯imu的标定吗，loader.cpp代码中odom坐标系是通过imu积分得到的吗，如果是，积分得到的漂移会不会很大？期待你的回复

[dezhi-l]

我用的是车载组合导航（rtk）和雷达之间的标定，没有用imu做积分。 如果是imu-lidar标定的话，绕8字imu积分误差估计会很大，标定精度无法保证 @huangwefeng121