Parking accuracy problem

[hzkkk1997]

When testing some algorithm functions of autoware in the real car, I have a question about how the algorithm restricts the accuracy of parking. It seems that I do not see relevant parameters in the algorithm of lane_driving scenario. It is possible to directly control the distance between goal_pose and base_link during parking within a certain value. Does anyone have relevant experience to share below?

[maxime-clem]

possible to directly control the distance between goal_pose and base_link

There is currently no mechanism to guarantee parking accuracy.
The mission planner will consider the goal reached based on its parameters (see https://autowarefoundation.github.io/autoware.universe/main/planning/autoware_mission_planner/#parameters). But even if the goal is not reach, there is no mechanism to fix the inaccurate parking.

[hzkkk1997]

@maxime-clem Thank you very much for your reply. May I ask if you will consider optimizing here in the future?
I found in the test that sometimes the distance between base_link and goal_pose is within 10cm, and sometimes it is about 40cm. Is this error a normal effect?

[maxime-clem]

As far as I know there is no plan to improve this feature.
High errors are currently caused by inaccuracy in the tracking of the trajectory by the vehicle. This can mainly be solved by improving the accuracy of the controller.
It is also possible to imagine features to replan the trajectory is case of the high error.

[hzkkk1997]

@maxime-clem Thank you for your reply, then I understand that to alleviate this error, it is mainly necessary to improve the tracking accuracy of the controller. The simple way is to adjust the parameters of the controller, and further, it may be necessary to optimize the algorithm?

[maxime-clem]

Yes tuning the control parameters is the easiest way to reduce the tracking errors.

[liuXinGangChina]

hi @hzkkk1997

Considering using autoware in real world parking，you will find autoware’s parking framework is similar to ADAS parking，including path search and trajectory following。 But they are different in details：

modern chasis provide distance control interface to Parking control module， in that case ESC will do vehicle motion control Precisely considering actuator control latency，ego dynamics properties and many other factors

Above is one of those. After all you can get a “not so bad” parking result if you take more time tunning with your chasis

happy hacking！

Xingang

[hzkkk1997]

@liuXinGangChina Thank you for your reply. As you said, ADAS performs more accurate parking control through ESC, right? If autoware's algorithm is used to improve parking accuracy, is it also necessary to optimize and adjust the chassis of the controlled vehicle?

[liuXinGangChina]

Hi @hzkkk1997
That‘s right ，consider localization， autoware can easily achieve cm level accuracy that is very important for a good adas parking system
In terms of chasis， cause ESC did a lot of calibration for driver comfortable things and precisely vehicle control

In conclusion，autoware do most of the job and cooperate with ESC，in that case comfortable and smart parking shall be done

further more， autoware is developed to handle more complex job，like AVP ，Let’s push autoware to its new limit together

happy hacking

Xingang