feature(freespace_planner): enhance freespace planner search algorithm and add reparking feature for improved parking accuracy #6320
Unanswered
YCLiu-TD
asked this question in
Feature requests
Replies: 1 comment 4 replies
-
|
Thank you for sharing your observations and results. Your enhancements look good and it would be great to add them to the |
Beta Was this translation helpful? Give feedback.
4 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Checklist
I've read the contribution guidelines.
I've searched other issues and no duplicate issues were found.
I'm convinced that this is not my fault but a bug.
Current Behavior
The current parking algorithm used by the Autoware freespace planner can constrain the planned yaw angle to within 3°. However, because this is purely a global‐planning result, the vehicle’s actual final orientation in PSIM simulations may still exceed that threshold. In five test runs, the final yaw error exceeded the 3° limit in four cases.
parking.exp.mp4
Demo video of the experiment scenario
The root cause is that the planned path still includes a rotational maneuver near the final goal. Although the path is kinematically feasible, it does not compensate for the execution error during actual vehicle movement. Additionally, the current design of the path tracking controller decouples steering and throttle control. As a result, the vehicle may be considered to have met the stopping condition when its position is close to the goal pose, even though the steering alignment has not yet been completed.
Expected Behavior
In PSIM parking scenarios, the vehicle’s yaw angle error relative to the user‑specified target pose is supposed to stay within 3°.
Actual Behavior
It often exceeds this 3° threshold, yet the system still deems the parking maneuver complete.
Proposed Enhancement
Refine the A* algorithm’s heuristic function so that the generated trajectory becomes progressively straighter as it approaches the final goal, reducing the controller’s tracking burden and thereby minimizing the final pose error.
Improved.mp4
In the demo with the adjusted algorithm, the final orientation error was reduced to 0.497° (the baseline average error over 10 parking tests in this scenario was 3.8°).
Emulate human behavior by adding a reparking feature: if the vehicle’s final parking orientation exceeds a specified deviation threshold, trigger a secondary parking maneuver that generates a combined pull‑out-and‑alignment trajectory to correct the angle. Allow up to three corrective attempts to iteratively reduce the error, while alleviating path constraints imposed by nearby obstacles.
reparking.webm
After executing the reparking function once to realign with the goal pose, the final orientation error was 0.9°.
Versions
OS: Ubuntu 22.04
ROS2: Humble
Autoware: Autoware.universe:main
Beta Was this translation helpful? Give feedback.
All reactions