Scenic crashes on custom map

[ipirwani-blip]

System Details

1. Python version: 3.10.19
2. Scenic version: 3.1.0
3. Operation system: Linux a990320-lcedt 6.17.0-14-generic #14~24.04.1-Ubuntu SMP PREEMPT_DYNAMIC Thu Jan 15 15:52:10 UTC 2 x86_64 x86_64 x86_64 GNU/Linux
4. Simulator: CARLA 0.9.16

Detailed Description

1. Scenic has a bug
2. scenic --2d --model scenic.simulators.carla.model --simulate --count 1 --seed 12345 /home/user/carla-camera-placement/monolith/client/test_data/rag18.scenic -b
3. Error log:

Using random seed = 12345
Beginning scenario construction...
/home/user/miniconda3/envs/carla-env/lib/python3.10/site-packages/pygame/pkgdata.py:25: UserWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for removal as early as 2025-11-30. Refrain from using this package or pin to Setuptools<81.
  from pkg_resources import resource_stream, resource_exists
pygame 2.6.1 (SDL 2.28.4, Python 3.10.19)
Hello from the pygame community. https://www.pygame.org/contribute.html
  Parsing OpenDRIVE file...
Traceback (most recent call last):
  File "/home/user/miniconda3/envs/carla-env/bin/scenic", line 3, in <module>
    from scenic.__main__ import dummy
  File "/home/user/Scenic_repo/src/scenic/__main__.py", line 195, in <module>
    scenario = errors.callBeginningScenicTrace(
  File "/home/user/Scenic_repo/src/scenic/core/errors.py", line 282, in callBeginningScenicTrace
    return func()
  File "/home/user/Scenic_repo/src/scenic/__main__.py", line 196, in <lambda>
    lambda: translator.scenarioFromFile(
  File "/home/user/Scenic_repo/src/scenic/syntax/translator.py", line 138, in scenarioFromFile
    return _scenarioFromStream(
  File "/home/user/Scenic_repo/src/scenic/syntax/translator.py", line 168, in _scenarioFromStream
    compileStream(stream, namespace, compileOptions, filename)
  File "/home/user/Scenic_repo/src/scenic/syntax/translator.py", line 317, in compileStream
    executeCodeIn(code, namespace)
  File "/home/user/Scenic_repo/src/scenic/syntax/translator.py", line 556, in executeCodeIn
    exec(code, namespace)
  File "/home/user/carla-camera-placement/monolith/client/test_data/rag18.scenic", line 1, in <module>
    """ TITLE: Nighttime Intersection Scenario
  File "/home/user/Scenic_repo/src/scenic/syntax/veneer.py", line 982, in model
    module = importlib.import_module(modelName)
  File "/home/user/miniconda3/envs/carla-env/lib/python3.10/importlib/__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "<frozen importlib._bootstrap>", line 1050, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1027, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1006, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 688, in _load_unlocked
  File "/home/user/Scenic_repo/src/scenic/syntax/translator.py", line 394, in exec_module
    code, pythonSource = compileStream(
  File "/home/user/Scenic_repo/src/scenic/syntax/translator.py", line 317, in compileStream
    executeCodeIn(code, namespace)
  File "/home/user/Scenic_repo/src/scenic/syntax/translator.py", line 556, in executeCodeIn
    exec(code, namespace)
  File "/home/user/Scenic_repo/src/scenic/simulators/carla/model.scenic", line 45, in <module>
    from scenic.domains.driving.model import *
  File "<frozen importlib._bootstrap>", line 1027, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1006, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 688, in _load_unlocked
  File "/home/user/Scenic_repo/src/scenic/syntax/translator.py", line 394, in exec_module
    code, pythonSource = compileStream(
  File "/home/user/Scenic_repo/src/scenic/syntax/translator.py", line 317, in compileStream
    executeCodeIn(code, namespace)
  File "/home/user/Scenic_repo/src/scenic/syntax/translator.py", line 556, in executeCodeIn
    exec(code, namespace)
  File "/home/user/Scenic_repo/src/scenic/domains/driving/model.scenic", line 74, in <module>
    network : Network = Network.fromFile(globalParameters.map, **globalParameters.map_options)
  File "/home/user/Scenic_repo/src/scenic/domains/driving/roads.py", line 1078, in fromFile
    network = handlers[ext](path, **kwargs)
  File "/home/user/Scenic_repo/src/scenic/domains/driving/roads.py", line 1120, in fromOpenDrive
    road_map.parse(path)
  File "/home/user/Scenic_repo/src/scenic/formats/opendrive/xodr_parser.py", line 1489, in parse
    raise ValueError(
ValueError: planView of road 7 has inconsistent length

Steps To Reproduce

1. Scenic file:

""" TITLE: Nighttime Intersection Scenario 
AUTHOR: [Your Name], [Your Email] 
DESCRIPTION: 20 cars approach a 4-way intersection at night with a fixed camera.
"""

#################################
# MAP AND MODEL
#################################
param map = localPath('/home/user/CARLA/CarlaUE4/Content/Carla/Maps/OpenDrive/HQ.xodr')
param carla_map = 'HQ'
model scenic.simulators.carla.model
import carla

from utils.scenic_utils import get_lanes_within_distance

models_with_lights = [
    'vehicle.dodge.charger_police_2020',
    'vehicle.dodge.charger_2020',
    'vehicle.ford.crown',
    'vehicle.lincoln.mkz_2020',
    'vehicle.mercedes.coupe_2020',
    'vehicle.mini.cooper_s_2021',
    'vehicle.nissan.patrol_2021',
#    'vehicle.carlamotors.firetruck',
    'vehicle.ford.ambulance',
#    'vehicle.mercedes.sprinter',
    'vehicle.volkswagen.t2_2021',
#    'vehicle.mitsubishi.fusorosa'
]

#################################
# CONSTANTS
#################################
MODEL = 'vehicle.lincoln.mkz_2017'
TERM_TIME = 60

weather_params = {
    'cloudiness': Range(0, 100).sample(),
    'precipitation': Range(0, 100).sample(),
    'sun_altitude_angle': TruncatedNormal(-60, 30, -180, 10).sample(), # Night-time
    'fog_density': Range(0, 100).sample(),
    'wetness': Range(0, 100).sample()
}

param weather = weather_params

#################################
# AGENT BEHAVIORS
#################################
behavior EgoBehavior():
    # Ensure lights are on for night-time
    take SetVehicleLightStateAction(carla.VehicleLightState.All)
    do AutopilotBehavior(target_speed=Range(2, 5))

#################################
# SPATIAL RELATIONS
#################################
# 1. Sample a concrete intersection
# Note: filter returns a generator, so we convert to a list
# Filter for 4-way intersections where all incoming lanes are sufficiently long
intersections = list(filter(lambda i: (i.is4Way or i.is3Way) and any(s.isTrafficLight for s in i.signals), network.intersections))
# Use .sample() to get a concrete Intersection object immediately
intersection = Uniform(*intersections).sample()

# 2. Get the center point (as a concrete coordinate)
# Use the centroid of the intersection's polygon and convert to a Scenic Vector
from scenic.core.vectors import Vector
intersectionCenter = Vector(*intersection.polygon.centroid.coords[0])

cars = []
#################################
# SCENARIO SPECIFICATION
#################################
# Compute valid lanes once before the loop
valid_lanes = get_lanes_within_distance(intersection.incomingLanes, 100, network)

for _ in range(30):
    # Since intersection is a concrete object, we can access its incomingLanes list
    # Filter lanes to only include those that are long enough
    lane = Uniform(*valid_lanes)
    
    # Spawn vehicles 5-10 meters away from the intersection
    # We use 'at a distance' from the intersection center
    spawn_pt = new OrientedPoint on lane.centerline
    
    car = new Car at spawn_pt, 
              with color Color.defaultCarColor(),
              with blueprint Uniform(*models_with_lights),
              with behavior EgoBehavior()
    require (distance from car to intersection) >= 5.0
    # Add a requirement that this car is separated from all previously spawned cars
    for other_car in cars:
        require (distance from car to other_car) >= 8.0
    cars.append(car)

terminate after TERM_TIME seconds

#################################
# WRITE INTERSECTION POINT AS COLLISION_LOCATION YAML
#################################
# Use the intersection center as the "location" in the same format as
# incident_point_sampler output (e.g. 1.location.yaml) for downstream
# camera placement / SDG pipeline.
import yaml
_cx, _cy = intersection.polygon.centroid.coords[0]
_cy = -_cy
_z = 0.0  # 2D map; use 0 for z
_output = {
    'meta_data': {
        'actor_1_location_at_collision': {'x': _cx, 'y': _cy, 'z': _z},
        'actor_2_location_at_collision': {'x': _cx, 'y': _cy, 'z': _z},
        'distance': 20.0,
        'fov': 110,
        'height': 7.0,
        'location': {'x': _cx, 'y': _cy, 'z': _z},
        'number_of_cameras': 5,
        'sampler_type': 'uniform_sampler',
    }
}
with open(localPath('rag18.location.yaml'), 'w') as _f:
    yaml.dump(_output, _f, default_flow_style=False, sort_keys=False)

#################################
# WRITE WEATHER PARAMS IN A WEATHER JSON FILE
#################################
import json
with open(localPath('rag18.weather.json'), 'w') as _f:
    json.dump(weather_params, _f, sort_keys=False, indent=2)

2. I can't attach the custom map; system wont let me do that. How can I communicate this?

Issue Submission Checklist

• I am reporting an issue, not asking a question
• I checked the open and closed issues, forum, etc. and have not found any solution
• I have provided all necessary code, etc. to reproduce the issue

[dfremont]

Hi Imran, could you please try putting the map file in a .zip and then dropping it into the comment box?

[ipirwani-blip]

@dfremont , you can try this:

1. Go to https://avesreality.com/synthetic-data/
2. Scroll down to "Want to Build Your Own Workflow? - Get Started Now!" and click on "Request Demo Data"
3. You'll need to take a brief survey to unlock a link to obtain the CARLA map as a package. In the "Area of Interest" type in "San Jose, CA, USA"

Let me know if it doesn't work for some reason.

[dfremont]

Hi Imran, thanks for the report! Lola fixed the first issue this map triggers but there are a couple more I'm still working on.

In the mean time, you might consider telling the maker of the map that it is violation of the OpenDRIVE spec in a couple of ways. First, section 9.2 states

The s-value of each <geometry> shall be the sum of all <geometry> lengths prior

The first two geometries of Road 7 have lengths 0.4512 and 0.7013, so the s-values should be 0, 0.4512, 1.1525. But they are actually 0, 0.4512, 1.1526. The numerical error is just barely large enough to exceed the fudge factor our parser allows (0.0001), which should be plenty according to OpenDRIVE section 6.1:

All floating-point numbers used in ASAM OpenDRIVE are IEEE 754 [12] double precision floating-point numbers. To ensure accurate representation of floating-point numbers in the XML representation, implementations should use a known correct accuracy preserving minimal floating-point printing algorithm (for example [13], [14]) or ensure that 17 significant decimal digits are always produced, for example using the "%.17g" ISO C printf modifier.

It looks like the generator for this map is rounding to 4 digits after the decimal point, which is not enough according to the above section.

The second issue is that in Road 151, second lane section, lane -3 has negative width at s=0, since the constant coefficient a is negative:

<width sOffset="0.0000" a="-2.2054036006746713e-05" b="0.0" c="0.005395891785627742" d="-0.00011221948179247459" />

This violates OpenDRIVE spec section 11.6.1:

Width (ds) shall be greater than or equal to zero.

We can of course make Scenic more liberal in what it accepts here (and we're doing so), but it would help us and other consumers of the map data if it followed the spec.

[ipirwani-blip]

Thanks, @dfremont and @lola831 for the fix and the detailed explanation. I'll pass the report on opendrive spec infractions to the map provider.

[dfremont]

Hi Imran, I think both maps will work with #468 (they seem to on my end) -- please give it a try and let us know if there are still problems.

Here are some issues with the montague.xodr map that I came across while working on this:

• There are geometries with length zero, e.g. the last geometry of Road 1102 (violating section 9.2 of the spec, which states that the length attribute must be strictly greater than zero). We already had a workaround for extremely short geometries but I hadn't seen one with exactly zero length before.
• There are many short spiral geometries with extremely fast fluctuations in curvature. Even some of the relatively less curved ones, like the first geometry of Road 1904 (curvature varying from -98/m to +103/m in 7 cm), cause the edges of the lanes to double back and intersect themselves. I put in a heuristic to skip such geometries but the lane shapes may get distorted, since some of these geometries have nontrivial length: the first geometry of Road 814 is one of these spirals and it is 57 cm long, 4% of the road's total length. Note that the curvature of Road 814 jumps from -20/m at the end of that spiral to 0.1/m at the start of the next geometry, so I'm pretty sure the map data is wrong.
• Junction 2199 seems to be missing a connection for Road 2200 (which claims to be part of that junction). The existing connections only connect the curbs: there is no connection for the actual driving lane (of Road 1067, the only incoming road). Since the junction contains no connections for drivable lanes, our library is ignoring it for now.

I also noticed that for some reason Scenic takes a very long time to pickle and unpickle the Montague map (i.e. save and load the .snet cached version), so that it's actually faster to load the .xodr file from scratch each time! That's never happened before and I plan to investigate, but in the mean time you can disable caching by adding something like this to your Scenic program:

param map_options = dict(writeCache=False)

Thanks again for the report -- this helped us fix several issues in our OpenDRIVE library that have been around for a long time but weren't triggered by most of the maps we've worked with before.