Lane detection and holding errors

[seitzseb]

Detailed Description

The agent struggles to maintain a steady lane position, leading to deviations that may intrude on adjacent lanes or off-road areas. Deviating off road is likely the source of further Problems that can be avoided.

Definition of Done

Agent maintains consistent lane position in all test cases.
Improved lane stability in various traffic conditions.

Effort Estimate

7

Testability

Test with lane-keeping scenarios, observing for any deviations.

Dependencies

Depends on perception and control adjustments.

[vinzenzm]

This is also somewhat a duplitcate of #358. Implementing some sort of lane detection for relative "positioning"/error correction seems however to be a good idea.

If this Comment receives 2 or more likes I will rename the issue to "Lane detection" only. Then we can use this Issue for brainstorming about how lane detection could be accomplished.

[vinzenzm]

One approach would be lane detection with "classical" vision and image processing.
This could be achieved with e.g. Canny Edges (OpenCV Library):
https://docs.opencv.org/4.x/da/d22/tutorial_py_canny.html
or HoughLines:
https://docs.opencv.org/3.4/d9/db0/tutorial_hough_lines.html

If we decide on implementing such feature we would also have to take into account how we can propagate detected lines through the planning stage. Is it sufficient to a; tell planning the deviation to the center of two lines we are in or b; we need to send planning a more complicated datastructure, maybe with other lanes we see