As group project at uni we have been assigned a project, in which we have to implement the same simple traffic scenario, an ego vehicle making a U-turn in some light traffic, using both SUMO and CARLA simulators independently then as a co-simulation, in which every vehicle are controlled by SUMO except the ego vehicle, which is controlled by CARLA. The main objective would be to log some performance metrics to compare the different simulations. We managed to make the independent simulations, but have trouble making doing the co-simulation.
We first made a scenario descriptor function to initialize both simulation, as follows:
class Actor:
sumo_index = 0
def __init__(self, pos: tuple, rot: tuple, destionation_pos: tuple, type: str, control: tuple, sensor: bool = False,
sens_rel_loc: tuple = None):
self.pos = pos
self.destination_pos = destionation_pos
self.rot = rot
self.type = type
self.sumo_id = f'vehicle{Actor.sumo_index}'
Actor.sumo_index += 1
self.control = control
self.sensor = sensor
self.sens_rel_loc = sens_rel_loc
def __str__(self):
print(
f'Actor type: {self.type}\n '
f'Actor position: x:{self.pos[0]},y:{self.pos[1]},z:{self.pos[2]}\n '
f'Actor rotation: pitch:{self.rot[0]},yaw:{self.rot[1]},roll:{self.rot[2]}\n '
f'Actor control: throttle:{self.control[0]},steer:{self.control[1]},brake:{self.control[2]}')
class ScenarioDescription:
ego_actor = Actor(
(-426.5, 30.4, 0.5),
(0, 0, 0),
(-438.57, 10.17, -0.00),
'vehicle.mercedes.coupe_2020',
(1.0, 0.0, 0.0),
sensor=True,
sens_rel_loc=(-5.7, 0, 3.7))
other_actors = [
Actor(
(-426.4, 26.9, 0.5),
(0, 0, 0),
(-326.72, 26.91, 0.22),
'vehicle.audi.a2',
(0.6, 0.0, 0.0)),
Actor(
(-447.2, 37.4, 0.5),
(0, 0, 0),
(-220.10, 37.37, 3.68),
'vehicle.tesla.model3',
(1.0, 0.0, 0.0)),
Actor(
(-298.2, 5.4, 3),
(0, 180, 0),
(-446.78, 6.99, -0.02),
'vehicle.audi.etron',
(1.0, 0.0, 0.0)),
Actor(
(-315.9, 12.7, 3),
(0, 180, 0),
(-464.65, 14.54, -0.01),
'vehicle.tesla.cybertruck',
(1.0, 0.0, 0.0)),
Actor(
(-347.7, 16.1, 3),
(0, 180, 0),
(-470.86, 17.06, -0.03),
'vehicle.dodge.charger_police',
(1.0, 0.0, 0.0)),
Actor(
(-384.5, -8.5, 3),
(0, 90, 0),
(-384.50, -8.50, -0.00),
'vehicle.ford.mustang',
(0.0, 0.0, 0.0)),
Actor(
(-380.7, -7.4, 3),
(0, 90, 0),
(-380.70, -7.40, 0.00),
'vehicle.audi.tt',
(0.0, 0.0, 0.0)),
]
all_actor = [ego_actor]
all_actor.extend(other_actors)
def conv2sumoLoc(loc: tuple) -> tuple:
sumo_loc = (loc[0] + 503.02, -loc[1] + 423.76)
return sumo_loc
Then we did the SUMO and CARLA:
SUMO
from __future__ import absolute_import
from __future__ import print_function
import os
import sys
from scenario_descriptor import ScenarioDescription as sc
from scenario_descriptor import conv2sumoLoc
# we need to import python modules from the $SUMO_HOME/tools directory
if 'SUMO_HOME' in os.environ:
tools = os.path.join(os.environ['SUMO_HOME'], 'tools')
sys.path.append(tools)
else:
sys.exit("please declare environment variable 'SUMO_HOME'")
from sumolib import checkBinary # noqa
import traci # noqa
def run():
step = 0
vehic_num = 0
start_edge_ego, _, start_lane_ego = traci.simulation.convertRoad(*conv2sumoLoc(sc.ego_actor.pos))
des_edge_ego, _, des_lane_ego = traci.simulation.convertRoad(*conv2sumoLoc(sc.ego_actor.destination_pos))
traci.route.add(f'trip{vehic_num}', [f'{start_edge_ego}', f'{des_edge_ego}'])
traci.vehicle.add(sc.ego_actor.sumo_id, f'trip{vehic_num}', typeID=sc.ego_actor.type.split('_')[0])
traci.vehicle.moveToXY(sc.ego_actor.sumo_id, f'{start_edge_ego}', f'{start_lane_ego}',
*conv2sumoLoc(sc.ego_actor.pos), keepRoute=1)
vehic_num += 1
for actor in sc.other_actors:
start_edge_actor, _, start_lane_actor = traci.simulation.convertRoad(*conv2sumoLoc(actor.pos))
des_edge_actor, _, des_lane_actor = traci.simulation.convertRoad(*conv2sumoLoc(actor.destination_pos))
traci.route.add(f'trip{vehic_num}', [f'{start_edge_actor}', f'{des_edge_actor}'])
traci.vehicle.add(actor.sumo_id, f'trip{vehic_num}', typeID=actor.type)
traci.vehicle.moveToXY(actor.sumo_id, f'{start_edge_actor}', f'{start_lane_actor}',
*conv2sumoLoc(actor.pos), keepRoute=1)
vehic_num += 1
while step < 1000:
traci.simulationStep()
step += 1
traci.close()
if __name__ == '__main__':
sumoBinary = checkBinary('sumo-gui')
traci.start([sumoBinary, '-c', 'D:\Carla_bullshit\WindowsNoEditor\Co-Simulation\Sumo\examples\Town04.sumocfg',
'--step-length', '0.023'])
run()
CARLA
import os
import time
import carla
import numpy as np
import pandas as pd
import pygame
from scenario_descriptor import ScenarioDescription as sc
CAMERA_WIDTH = 1280
CAMERA_HEIGHT = 720
# Pygame Setup
pygame.init()
screen = pygame.display.set_mode((CAMERA_WIDTH, CAMERA_HEIGHT))
actor_list = []
metrics_df = pd.DataFrame(
columns=['actor_type', 'sumo_id', 'velocity', 'travelled_distance', 'throttle', 'steer', 'brake', 'acceleration',
'acc_x', 'acc_y', 'loc_x', 'loc_y', 'heading', 'time_step'])
def visualize_image(image):
image.convert(carla.ColorConverter.Raw)
array = np.frombuffer(image.raw_data, dtype=np.dtype("uint8"))
array = np.reshape(array, (image.height, image.width, 4))
array = array[:, :, :3]
array = array[:, :, ::-1]
surface = pygame.surfarray.make_surface(array.swapaxes(0, 1))
screen.blit(surface, (0, 0))
def log_metrics(path: str = None, last_tick = False):
global metrics_df
if path is None:
path = os.getcwd()
path = os.path.join(path, 'logs')
if not os.path.exists(path):
os.mkdir(path)
filepath = os.path.join(path, f'logged_metrics_{time.time()}.csv')
for ac in actor_list:
sc_actor = [sc_ac for sc_ac in sc.all_actor if sc_ac.type == ac.type_id][0]
actor_type = ac.type_id
sumo_id = sc_actor.sumo_id
vel = ac.get_velocity().length()
time_step = world.get_snapshot().timestamp.delta_seconds
travelled_dis = vel * time_step
try:
type_df = metrics_df.loc[metrics_df['actor_type'] == ac.type_id]
travelled_dis = travelled_dis + type_df.iloc[-1,3]
except IndexError:
pass
control = ac.get_control()
throttle = control.throttle
steer = control.steer
brake = control.brake
acc_x = ac.get_acceleration().x
acc_y = ac.get_acceleration().y
acceleration = (acc_x ** 2 + acc_y ** 2) ** (1 / 2)
loc_x = ac.get_location().x
loc_y = ac.get_location().y
heading = ac.get_transform().rotation.yaw
new_row = {'actor_type': actor_type, 'sumo_id': sumo_id, 'velocity': vel, 'travelled_distance': travelled_dis,
'throttle': throttle, 'steer': steer, 'brake': brake, 'acceleration': acceleration, 'acc_x': acc_x,
'acc_y': acc_y,'loc_x':loc_x,'loc_y':loc_y,'heading':heading,'time_step':time_step }
metrics_df = metrics_df.append(new_row,ignore_index=True)
if last_tick:
metrics_df.to_csv(filepath,index=False)
try:
# Client
client = carla.Client('localhost', 2000)
client.set_timeout(10.0)
client.load_world('Town04')
# World
world = client.get_world()
default_settings = world.get_settings()
settings = world.get_settings()
settings.synchronous_mode = True
settings.fixed_delta_seconds = 0.023
world.apply_settings(settings)
world.set_weather(carla.WeatherParameters.ClearNoon)
blueprint_library = world.get_blueprint_library()
# Ego vehicle with camera attached to it's back
vehicle_bp = blueprint_library.find(sc.ego_actor.type)
camera_bp = blueprint_library.find('sensor.camera.rgb')
camera_bp.set_attribute('image_size_x', str(CAMERA_WIDTH))
camera_bp.set_attribute('image_size_y', str(CAMERA_HEIGHT))
ego_vehicle_location = carla.Transform(carla.Location(*sc.ego_actor.pos), carla.Rotation(*sc.ego_actor.rot))
ego_vehicle = world.try_spawn_actor(vehicle_bp, ego_vehicle_location)
ego_vehicle.apply_control(carla.VehicleControl(*sc.ego_actor.control))
camera_relative_loc = carla.Transform(carla.Location(*sc.ego_actor.sens_rel_loc), carla.Rotation(pitch=-15))
camera = world.spawn_actor(camera_bp, camera_relative_loc, attach_to=ego_vehicle)
camera.listen(lambda data: visualize_image(data))
actor_list.append(ego_vehicle)
for actor in sc.other_actors:
vehicle_bp = blueprint_library.find(actor.type)
vehicle_location = carla.Transform(carla.Location(*actor.pos), carla.Rotation(*actor.rot))
vehicle = world.try_spawn_actor(vehicle_bp, vehicle_location)
vehicle.apply_control(carla.VehicleControl(*actor.control))
actor_list.append(vehicle)
world.tick()
# Spectator
spectator = world.get_spectator()
spectator_location = carla.Transform(
carla.Location(x=-445.9, y=24.23, z=44.5), carla.Rotation(pitch=-47.0, yaw=0.0, roll=0.0))
spectator.set_transform(spectator_location)
running = True
screen.fill((0, 0, 0))
pygame.display.flip()
distance = 0.0
while running:
for event in pygame.event.get():
if event.type == pygame.QUIT:
running = False
world.tick()
current_location = ego_vehicle.get_transform().location
distance += ego_vehicle_location.location.distance(current_location)
ego_vehicle_location.location = current_location
# set Traffic Lights
actors = world.get_actors().filter('traffic.traffic_light')
for actor in actors:
state = carla.TrafficLightState.Red
light_dis = actor.get_location().distance(current_location)
if light_dis < 55:
state = carla.TrafficLightState.Yellow
if light_dis < 45:
state = carla.TrafficLightState.Green
actor.set_state(state)
if distance > 20:
ego_vehicle.apply_control(carla.VehicleControl(brake=0.2, steer=0.0))
[a for a in actor_list if a.type_id == 'vehicle.audi.a2'][0].apply_control(
carla.VehicleControl(throttle=1.0, steer=0.0))
if distance > 34:
ego_vehicle.apply_control(carla.VehicleControl(throttle=1.0, steer=-0.3))
if distance > 67:
ego_vehicle.apply_control(carla.VehicleControl(throttle=4.0, steer=0.0))
if distance > 83:
ego_vehicle.apply_control(carla.VehicleControl(throttle=1.0, steer=0.07))
if distance > 87:
ego_vehicle.apply_control(carla.VehicleControl(throttle=1.0, steer=0.0))
pygame.display.flip()
log_metrics(last_tick=not running)
finally:
world.apply_settings(default_settings)
world.tick()
for actor in actor_list:
actor.destroy()
camera.destroy()
pygame.quit()
Both work just fine, but we have no clue how to stitch them together to make the co-simulation, note that we had no lectures nor practice on the software to teach us how to use them so we were figuring out as we went. Unfortunately we are now stuck, please help us.
