waypoint.next()/waypoint.previous() issue when connecting two lanes in opposite direction

[matejm42]

CARLA version: 0.9.15
Platform/OS: Ubuntu 20.04
Problem you have experienced: When calling wp.next(150) it returns a list with a waypoint at the same position as wp
What you expected to happen: Return the waypoint 150 meters from the wp or an empty list []
Steps to reproduce: Call wp.next(150) from road 25, lane -8 on map Town06

Code that causes the bug

import carla
adress = "127.0.0.1"
port = 2000

client = carla.Client(adress, port)
client.set_timeout(30)
world = client.load_world("Town06")
map = world.get_map()


wpev: carla.Waypoint = map.get_waypoint(
                carla.Location(215, 54, 0),
                project_to_road=True,
                lane_type=carla.LaneType.Driving)

wpcurb = wpev.get_right_lane().get_right_lane()
print(wpcurb.transform.location, f"{wpcurb.road_id=}, {wpcurb.lane_id=}")

wpnext1 = wpcurb.next(1)[0]
print(wpnext1.transform.location, f"{wpnext1.road_id=}, {wpnext1.lane_id=}")

wpnext2 = wpcurb.next(12)[0]
print(wpnext2.transform.location, f"{wpnext2.road_id=}, {wpnext2.lane_id=}")


wpnext3 = wpcurb.next(14)[0]
print(wpnext3.transform.location, f"{wpnext3.road_id=}, {wpnext3.lane_id=}")

wpnext4 = wpcurb.next(16)[0]
print(wpnext4.transform.location, f"{wpnext4.road_id=}, {wpnext4.lane_id=}")

wpnext5 = wpcurb.next(152.188)[0]
print(wpnext5.transform.location, f"{wpnext5.road_id=}, {wpnext5.lane_id=}")

Location(x=214.900406, y=54.925457, z=0.000000) wpcurb.road_id=25, wpcurb.lane_id=-8
Location(x=215.930038, y=54.931492, z=0.000000) wpnext1.road_id=25, wpnext1.lane_id=-8
Location(x=221.930862, y=55.283535, z=0.000000) wpnext2.road_id=745, wpnext2.lane_id=1
Location(x=220.301590, y=54.936768, z=0.000000) wpnext3.road_id=745, wpnext3.lane_id=1
Location(x=200.564636, y=54.777279, z=0.000000) wpnext4.road_id=25, wpnext4.lane_id=-8
Location(x=214.900421, y=54.925457, z=0.000000) wpnext5.road_id=25, wpnext5.lane_id=-8

Images of the road

[matejm42]

Another issue that I encounter is that the simulator crashes every second time I run the enclosed code.

[matejm42]

I suggested a solution in #7137

Waypoint.next(distance) is in the cpp source as Map::GetNext. This function computes distance to the end of the road section and returns either list with waypoint pointing to the same line, or recursively calls itself to the successors with the remaining distance.

Successors are obtained from the Map::GetSuccessors which calls GetNextLanes.
Lanes are defined in MapBuilder::GetLaneNext.

The critical part for us is this:

// successor and predecessor (road and lane)
    LaneId next;
    RoadId next_road;
    if (lane_id <= 0) {
      next_road = road.GetSuccessor();
      next = lane->GetSuccessor();
    } else {
      next_road = road.GetPredecessor();
      next = lane->GetPredecessor();
    }

In the described case, lane -8 on road 25 is negative, so next() returns its successor - lane 1 on road 745. Lane 1 is positive, so next() returns its predecessor, lane -8 on road 25, and we have a loop.

This sounds like an issue in the map that could be fixed by changing the id of the lane on road 745 from 1 to -1, but that would lead to the same issue when going from the second side - road 45.

Road 745 connects roads 25 and 45. When calling wp.next() from lane 1 on road 45, the lane id is positive, so it returns its predecessor, lane 1 on road 745. Calling another next() gets the predecessor of road 745, which is road 25. That is right. It will return us to the wrong end of the road 25 anyway, though.

This part of the map is problematic because the road 745 is joining the road 25 and is never supposed to be used in the opposite direction. That is why everything works fine for the driving lanes, where 745 is not marked as the successor of 25. The problem is for the side 'shoulder' lane representing a curb that connects two lanes that are never meant to be connected from a driving perspective.

Our solution for this problem, suggested in #7137, is to check if the successor can cause the loop and skip it in that case.