Wrong initial heading in Autoware initialization

[chivas1000]

Hi and I’m using Autoware.Universe Galactic branch and using the GNSS Poser to convert navsatfix message to MGRS local cartesian pose as pose with covariance to autoware as pose for initialization (not yet for NDT pose regularization, which are set to false)

but the problem is that, it seems that the GNSS pose came in the autoware, but since I set the “use autoware orientation” to false(since I didn’t know which sensor data should be put in this message), so GNSS calculates the differential position as pose, however, the pose moving behavior in autoware rviz seem to be totally reverse to what the car actual does, which means that when the car move forward and turn left, it go back and turn right, in which I think its initial heading are opposite to what it actually does.

How do I solve the problem?

PS: I have to notice that I am using a different IMU unit as the IMU input to pose estimator(which might not align with the global heading of the GNSS unit, but the GNSS are came from a Inertial navigation system unit, so it can also provide 6 axis data but in lower accuracy and frequency), Do I need to switch IMU input to INS?

PS2: the autoware documentation said that if I need to use GNSS, then I have to make the HD MAP to be in global coordinates to use GNSS data for localization, but it does not tell how to build MAP in global coordinates. The problem might be the HD MAP without GNSS's heading(coordinates or orientation) are not align with pose given from GNSS poser?

PS3: I am using modified version of sample_sensor_kit as the car model, but it is not offering gnss_base_link, so I’ve directly set base_link as “gnss_base_frame” and gnss_link are same orientation as sensor_kit_base_link, does changing the transform of gnss link help?

Thanks for reading my issue and any assistance would be greatly appreciated!

updates:
I here would be the map I was using, in which it not indicates the true East North Up:

And I shall managed to build a map with global coordinates, in which the red axis(x) are true East, green axis(y) are true north.

[YamatoAndo]

@chivas1000 Hi, from reading your comments, I thought it might be because the map is not in East North Up.

And I shall managed to build a map with global coordinates, in which the red axis(x) are true East, green axis(y) are true north.

After correcting the map, is the behavior of the vehicle still reversed?

[YamatoAndo]

If possible, sharing the rosbag and map could help resolve the issue more smoothly.

[chivas1000]

Hi @YamatoAndo,
sorry for late reply, thank you for your answer!
after using the map with global coordinates, the gnss location aligned with the map.

but I want to mention that, since the GNSS Poser are publishing the map to gnss_base_frame TF(I've set it to base_link), then it competes/ with NDT or some other pose estimator, I need to comment out
line 376 tf2_broadcaster_.sendTransform(transform_stamped);
https://github.com/autowarefoundation/autoware.universe/blob/galactic/sensing/gnss_poser/src/gnss_poser_core.cpp

also I suggest carefully calibrate the sensor kit and use the INS to give AutowareOrientation for those who wants to use GNSS/INS to give orientation in Autoware.

[YamatoAndo]

I'm glad to hear that it worked out well.

it competes/ with NDT or some other pose estimator,

The map to base_link TF is output by the ekf_localizer node.

[marioney]

I found this issue too. The tf published by gnss_poser competes with the one from ndt and it creates a wrong tf structure tree as base_link has two parents.

@YamatoAndo not sure about this, as @chivas1000 commented, the gnns_poser does publish this tf.

@chivas1000 have you solved this inconsistency ?

[miku54]

Hi, I am also using GNSS for initialization positioning, but I encountered some problems and I am not sure whether GNSS initialization is successful. I am very happy to see this post because it is a successful case. May I ask which parameter you asked in your question, "use autoware orientation" and "not yet for NDT pose regularization, which are set to false", do you need to change it to false? Which parameter is it? I did not modify this step, I suspect that I did not set the parameters correctly. Looking forward to your reply.

[YamatoAndo]

@miku54 Could you create a new discussion?
It would be helpful if you could provide detailed information, the version of Autoware, terminal logs (~/.ros/log//launch.log), screenshots from RViz, pcd map and rosbag.

[miku54]

@miku54 Could you create a new discussion? It would be helpful if you could provide detailed information, the version of Autoware, terminal logs (~/.ros/log//launch.log), screenshots from RViz, pcd map and rosbag.

Thanks for your reply. In order to facilitate the problem analysis, what data should be included in the rosbag? In addition to the point cloud data, GNSS, IMU, and ODOM sensor data, what other data should be provided? I will upload the data after preparing it.

[YamatoAndo]

At a minimum, sensor data and /tf_static are sufficient, but since we will filter unnecessary topics on our end, it’s fine if other topics are included as well.

[miku54]

NDT pose regularization

I want to know if I plan to init by gnss, do I need to set NDT pose regularization to true?

[YamatoAndo]

It is not necessary for estimating the initial position using GNSS.
Whether it is set to true or false does not affect the initial position estimation.
However, it might have an impact if you perform the initial position estimation again.