Test highway localization feasibility of Current Autoware Pipe-Line

[liuXinGangChina]

Checklist

• I've read the contribution guidelines.
• I've searched other issues and no duplicate issues were found.
• I've agreed with the maintainers that I can plan this task.

Description

Several months ago, we have test the ndt based localization pipeline under 60 km/h, you can find the result here
We also notice that autoware introduce a new localization vision based theory called "yabloc", together with test report here and here under 30km/h
In the great march to achieve L4 autonomous driving, it is necessary for autoware to fill the absence of highway secenario. In order to make it happen, we plan to focous on highway localization first.

Purpose

Test autoware's current localization pipeline with highway secenario
Leave comment for localization enhancement if necessary

Possible approaches

• 1. Scan a map of high-way test field
• 2. Integate our sensor into autoware
• 3. Perform road test for 4 velocity zones (70 - 100)
• 4. Perform road test for 4 velocity zones (70 - 100), with ramp for drive in or out of high way
• 5. Compare different localization pipe-line result and leave our comment for localization enhancement for highway scenario

Definition of done

1. integration finish
2. test report
3. test report with ramp secenario
4. comment for enhancement

[liuXinGangChina]

Morning Kento-san @KYabuuchi , currently we are working on creat the map for localization test,
Due to the difference sensor-configuration between tire4 and autocore, I wonder whether a 2Mp120°-fov camera suit your algorithm-yabloc ?

[KYabuuchi]

@liuXinGangChina Good morning. 2MP and 120° FoV are sufficient for operating YabLoc. 👍 (Increasing the resolution beyond this won't bring any benefits. )
Please note that YabLoc relies not only on the camera but also on GNSS, IMU, and vehicle wheel odometry.

[KYabuuchi]

By the way, the link in the initial post might be incorrect. Please check it.

you can find the result here

[liuXinGangChina]

By the way, the link in the initial post might be incorrect. Please check it.

you can find the result here

already update the link， thank you for remind

[liuXinGangChina]

@liuXinGangChina Good morning. 2MP and 120° FoV are sufficient for operating YabLoc. 👍 (Increasing the resolution beyond this won't bring any benefits. ) Please note that YabLoc relies not only on the camera but also on GNSS, IMU, and vehicle wheel odometry.

that will be great，since our camera meet yabloc's resuirement，and we have all the other sensors you mentioned， we will continue this task

thank you

[liuXinGangChina]

Morning，yabuuchi-san @KYabuuchi ，during the preparation of the test， we found a note that said “If the road boundary or road surface markings are not included in the Lanelet2, the estimation is likely to fail.” in the limitation notification of the code。

currently our test lanelet2 map only contain lane line of the road，can you provide some additional material or an example to tell us what formation “road boundary or road surface markings” should be like in a lanelet2 file

[KYabuuchi]

Hi @liuXinGangChina , "road boundary or road surface markings" includes lane lines, stop lines, crosswalks, bus stops, etc.

The figure below is a lanelet2 provided in the AWSIM tutorial, which is ideal as it contains all of crosswalk, stop lines, bus stops.

Since highways usually only has lane lines, I think your map is sufficiently compatible for YabLoc.

[liuXinGangChina]

Hi @liuXinGangChina , "road boundary or road surface markings" includes lane lines, stop lines, crosswalks, bus stops, etc.

The figure below is a lanelet2 provided in the AWSIM tutorial, which is ideal as it contains all of crosswalk, stop lines, bus stops.

Since highways usually only has lane lines, I think your map is sufficiently compatible for YabLoc.

got it，that‘s true. there are only lane line on the high way

[liuXinGangChina]

Hi, Yabuuchi-san @KYabuuchi . During our test on highway, There are some issue confuse me. In the image below we can see line-extract and graph-seg went well on our dataset, but we got nothing in "pf/match_image" and no lane projected to the /pf/lanelet2_overlay_image. Do you have any clue?

[KYabuuchi]

@liuXinGangChina Please check /localization/pose_estimator/yabloc/image_processing/projected_image & /localization/pose_estimator/yabloc/pf/cost_map_image are published correctly.

yabloc#image-topics-for-debug

projected_image is an image of the line segments and segmented results projected onto the ground.
cost_map_image is an image of the cost map generated from lanelet2.

If projected_image is not being published, there may be no tf from base_link to camera.
If cost_map_image is not being published, the some of lanelet2 elements might not be loaded properly.
Edit this as necessary:
https://github.com/autowarefoundation/autoware_launch/blob/main/autoware_launch/config/localization/yabloc/ll2_decomposer.param.yaml#L3