Give Town10HD_Opt's lane boundaries the markings CARLA draws (#6)

- Give Town10HD_Opt's lane boundaries the markings CARLA draws (f33ab82b84386489a7d714f0d767c0ce9f7d67c9)

README.md

@@ -128,18 +128,26 @@ That is 30 traffic-light regulatory elements for CARLA's 15 signals, one per lan
 </details>
 
 <details>
-<summary>Lane changes need tagged lane boundaries</summary>
+<summary>Lane markings</summary>
 
 The published lanelet2 maps carry geometry only: their boundary ways have no `type`. Lanelet2's traffic rules read the shared boundary to decide whether a vehicle may cross it, and an untagged boundary is one it may not. The routing graph then holds `adjacentLeft` / `adjacentRight` relations but no `left` / `right` ones, Autoware's route handler finds no lane change target, and the ego drives to the end of its own lane whatever the route says.
 
-The map here keeps the published geometry, so it has this property. Tagging the dividers between same-direction lanes as `line_thin` / `dashed` is what makes crossing legal. Tag only the lanes a scenario drives: these towns are built from thousands of very short lanelets, and with lane changes legal everywhere the behavior planner has been seen to spin without ever publishing a path.
+The boundaries here carry the markings CARLA draws, read by [`tag_lane_boundaries.py`](https://gist.github.com/hakuturu583/f11582b8ca65a5e2efae9fe884e1dfc0) from the CARLA waypoints at a quarter, half and three quarters of each lanelet outside a junction:
+
+| CARLA marking | lanelet2 `type` / `subtype` | boundaries | a lane change across it |
+| --- | --- | --- | --- |
+| Broken | `line_thin` / `dashed` | 38 | allowed |
+| Solid | `line_thin` / `solid` | 48 | not allowed |
+| SolidSolid | `line_thin` / `solid_solid` | 36 | not allowed |
+
+So a lane change is legal exactly where CARLA paints a broken line -- between same-direction lanes, on 38 lanelets each way -- and nowhere else: the centre lines are solid, and the boundaries inside junctions are left untyped, as is a road edge CARLA draws no line on. Where two readings of one boundary differ, the more restrictive wins.
 
 </details>
 
 <details>
 <summary>Provenance</summary>
 
-`lanelet2_map.osm` started as a verbatim copy of [`maps/vector_maps/lanelet2/Town10HD.osm`](https://bitbucket.org/carla-simulator/autoware-contents/src/master/maps/vector_maps/lanelet2/Town10HD.osm) in [carla-simulator/autoware-contents](https://bitbucket.org/carla-simulator/autoware-contents), with SHA-256 `cdd0a28664bcf9b5bd8867584a6fa6a02fa0f75a7c3dbb901ca216128ef86a17`. Two things have been added to it, and nothing of the original geometry changed: the `turn_direction` tags of its junction lanelets, read from CARLA's OpenDRIVE, and the traffic lights, stop signs and stop lines above, written by [`add_regulatory_elements.py`](https://gist.github.com/hakuturu583/f11582b8ca65a5e2efae9fe884e1dfc0) from a CARLA 0.10.0 server. `map_projector_info.yaml` states the projector that file is drawn against.
+`lanelet2_map.osm` started as a verbatim copy of [`maps/vector_maps/lanelet2/Town10HD.osm`](https://bitbucket.org/carla-simulator/autoware-contents/src/master/maps/vector_maps/lanelet2/Town10HD.osm) in [carla-simulator/autoware-contents](https://bitbucket.org/carla-simulator/autoware-contents), with SHA-256 `cdd0a28664bcf9b5bd8867584a6fa6a02fa0f75a7c3dbb901ca216128ef86a17`. Three things have been added to it, and nothing of the original geometry changed: the `turn_direction` tags of its junction lanelets, read from CARLA's OpenDRIVE; the lane markings above, written by [`tag_lane_boundaries.py`](https://gist.github.com/hakuturu583/f11582b8ca65a5e2efae9fe884e1dfc0) from a CARLA 0.10.0 server; and the traffic lights, stop signs and stop lines above, written by [`add_regulatory_elements.py`](https://gist.github.com/hakuturu583/f11582b8ca65a5e2efae9fe884e1dfc0) from a CARLA 0.10.0 server. `map_projector_info.yaml` states the projector that file is drawn against.
 
 `pointcloud_map.pcd` was recorded from a running CARLA 0.10.0 server. A LiDAR with no vehicle attached is placed at every lane sample of the world, 8 m apart and 2.0 m above the road, with an 80 m range. The returns are accumulated in the Autoware map frame and voxel-downsampled at 0.2 m. The recorder is `docker/devcontainer/carla/build-pointcloud-map.py` in [autowarefoundation/autoware](https://github.com/autowarefoundation/autoware).
 

autoware_maps/Town10HD_Opt/lanelet2_map.osm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0ec0f990b37f629a54f53c420b351607b737f44b819a4cfafc6bc172d60d6288
-size 2418320
+oid sha256:61458e39ba6c5f097c5c658d402d9925af6b94302802cc4818df775212a69e66
+size 2426748