search

carla-simulator / carla

Open-source simulator for autonomous driving research.
http://carla.org
MIT License
11.53k stars 3.72k forks source link

Problem with Map.get_waypoint() in combination with OpenDRIVE spiral #2855

Closed alexk-92580 closed 4 years ago

alexk-92580 commented 4 years ago

Hi CARLA Team,

we would like to report a problem we observed in the Map.get_waypoint() function in the Python API in CARLA version 0.9.9, when using it on a spiral geometry within a standalone OpenDRIVE .xodr map loaded via config.py -x.

We are using Map.get_waypoint() with project_to_road = True, lane_type_flag is default. Sporadically, when calling Map.get_waypoint() with the position of the ego vehicle as argument, the returned waypoint is not on the lane the vehicle is in, as expected, but instead on the adjacent left lane. The problem does consistently occur at the same location in the map, which is inside a spiral geometry defined in the OpenDRIVE map.

When inspecting the lane_id of the preceding correct waypoint and the first wrong waypoint, the lane_id changes to the value of the adjacent lane, although the vehicle clearly is close to the center of its original lane.

We appreciate your help! Alex

germanros1987 commented 4 years ago

@Axel1092 please take a look at this.

Axel1092 commented 4 years ago

Hi @alexk-92580, Can you please share more details about this issue? Does this happen within a junction? Does the spiral have a high curvature?

alexk-92580 commented 4 years ago

Hi @Axel1092, thanks for taking care of this! The problem does not occur within a junction. Also, the curvatures are relatively low. We did observe, that the problem does not occur on all lanes. The spiral has 3 driveable lanes, and we have the issue on the rightmost driveable lane (ID -4) within a spiral with the following openDRIVE definition around 1629 meters into the segment:

 <geometry s="2500.0" x="2455.725103068875" y="-1198.2529509679275" hdg="-1.0" length="4981.650000000001">
                <spiral curvStart="0.0005" curvEnd="-0.0009860953494204312"/>
  </geometry>

If you need further information, please let me know!

Axel1092 commented 4 years ago

The problem is probably due to the discretization of the road in segments that enable the spatial queries and might be not fine enough. It would be great if you can provide a snippet of the geometry record of this spiral (or just its length, curvStart and curvEnd) to do some testing.

alexk-92580 commented 4 years ago

Hi @Axel1092 , sorry, I forgot to add the code tags around the openDRIVE snippet in my previous comment. The geometry record of the particular spiral in question is:

<geometry s="2500.0" x="2455.725103068875" y="-1198.2529509679275" hdg="-1.0" length="4981.650000000001">
                <spiral curvStart="0.0005" curvEnd="-0.0009860953494204312"/>
  </geometry>

Let me know if you need any more information!

Axel1092 commented 4 years ago

Hi @alexk-92580, I found the problem, it was related to how we discretize the road in straight segment that was producing precision errors. A fix is placed in the related PR and will be included in the next release. Thank you for noticing this problem.