carla: Synchronous mode - tick hangs

Intro

I am running a simple script in synchronous mode which does the following:

• spawn an ego vehicle
• in a loop do:
  • set velocity
  • initiate tick
  • wait to receive data
  • store it.

Here is the code (i’ve simplified the synchronous mode example):

import carla
import queue


snapshots = queue.Queue()


def push_snapshot(snapshot):
    snapshots.put(snapshot)


def get_position(frame_id, vehicle_id):
    snapshot = snapshots.get(timeout=2)
    assert(frame_id == snapshot.frame)
    transform = snapshot.find(vehicle_id).get_transform()
    return transform.location.x, transform.location.y, transform.location.z


def main():
    sim_time = 0.0
    sim_time_end = 20.0
    delta_time = 0.05
    results = []

    client = carla.Client("127.0.0.1", 2000)
    world = client.get_world()

    # set mode to synchronous
    frame_id = world.apply_settings(carla.WorldSettings(True, False, delta_time))
    print("Synchronous mode will be applied in frame: {}".format(frame_id))

    # spawn car
    spectator = world.get_spectator()
    blueprint_library = world.get_blueprint_library()
    map_ = world.get_map()
    vehicles = blueprint_library.filter("vehicle")
    bp = vehicles[0]
    vehicle = world.spawn_actor(bp, map_.get_spawn_points()[0])
    world.tick()    # just so that I see it in the simulator

    world.on_tick(push_snapshot)

    cmd = [carla.command.ApplyVelocity(vehicle.id, carla.Vector3D(y=2.0))]    # constant for the minimal example
    while sim_time <= sim_time_end:
        client.apply_batch_sync(cmd)
        print("before tick")
        frame_id = world.tick()   # on newer versions raises an Exception from which you can't rly recover
        print("after tick {}".format(frame_id))
        sim_time += 0.05
        results.append((sim_time, get_position(frame_id, vehicle.id)))

    world.apply_settings(carla.WorldSettings(False, False, 0))
    vehicle.destroy()


if __name__ == '__main__':
    main()

Problem

The problem happens when running this multiple times, sometimes the tick call will just hang. I originally encountered this with the C++ API (0.9.7). After that, I was able to reproduce it with the PythonAPI on 0.9.7, 0.9.8, 0.9.9 as well.

In case of 0.9.7, the script would just hang forever. With the newer versions, an exception would be thrown because of the recently added tick timeout (#2556). But in my opinion this just hides the root problem.

In my opinion this is really a big shortcoming of Carla and the synchronous mode, and it basically makes the synchronous mode not that usable. If you want to have reproducible results, it is not possible as the frame (corresponding to the “hanging” tick call) would be dropped. The exception being thrown in newer versions due to a timeout is also not useful as you can’t really recover from it.

I’ve already found a few related issues:

• #2556
• #2070 - sadly closed, but I can see that the author and @nsubiron also commented about the same problem. It seems like the issue was forgotten.
• #1803
• #2581
• #2038

So here are my questions:

1. Do you agree with me that this is a problem (and can be reproduced)?
2. Is anything already planned regarding the fix (as I can see in the linked issue that @nsubiron also suggests getting rid of the cause of the problem instead of applying workarounds?

Additional info

IMPORTANT: The described problem doesn’t occur all the time. You might have to run the script multiple times to see it. I typically run it like:

FOR /L %v IN (1,1,50) DO python tick_hangs.py

Environment

OS: Win 10 Python: 3.7 Carla versions:

• 0.9.7 (locally built from 1f7669e4b308aa8e8f7612dde6666ba6eda14e9f) - tried both the python and C++ API
• 0.9.8 (official package) - tried only the python API
• 0.9.9 (official package) - tried only the python API