Carla Simulator Features

Overview

DI-drive defines simulator instance to unify the interfaces and standardize sub-modules. As a result, users can customize their simulator setting with a config dictionary, and get expected data by calling the instance’s methods. DI-drive currently uses Carla version higher than 0.9.9 for simulating Autonomous Driving. The Interaction between Carla server and DI-drive is proceeded through CarlaSimulator. The simulator contains a world map with roads, buildings and weather. The simulator has a hero vehicle for the user to control and get observations. The observation may come from a sensor created in the simulator associated to hero vehicle, a planner in the simulator to get target waypoints and select road options, and other running status returned by methods.

Carla Simulator Configuring

All the settings of simulator can be customized by a config dict. Here we show the default setting of the simulator and explain its contents.

config = dict(
    town='Town01',
    weather='random',
    sync_mode=True,
    delta_seconds=0.1,
    no_rendering=False,
    auto_pilot=False,
    n_vehicles=0,
    n_pedestrians=0,
    disable_two_wheels=False,
    col_threshold=400,
    waypoint_num=20,
    obs=list(),
    planner=dict(),
    aug=None,
    verbose=True,
    debug=False,
)

Note

Some of the configures can be changed by provided arguments when running init method, in order to make difference among episodes.

Common configurations

Here we explain some common configurations about world map and npc settings in simulator, the instruction of ‘planner’, ‘obs’ and ‘aug’ are shown next.

• town

  Name of the town map used in the simulator.

• weather

  Weather setting in the simulator. List as followings. If set to ‘random’, the simulator will randomly choose one.

  Value	Weather
  1	Clear Noon
  2	Cloudy Noon
  3	Wet Noon
  4	Wet Cloudy Noon
  5	Mid Rainy Noon
  6	Hard Rain Noon
  7	Soft Rain Noon
  8	Clear Sunset
  9	Cloudy Sunset
  10	Wet Sunset
  11	Wet Cloudy Sunset
  12	Mid Rain Sunset
  13	Hard Rain Sunset
  14	Soft Rain Sunset

• sync_mode

  Whether to run the simulator in synchronous mode. It is suggested to set to TRUE throughout the simulation.

• delta_seconds

  Time step in seconds between two frames. Only make sense in sync mode. We suggest to remain same throughout the simulation.

• no_rendering

  Whether to use no rendering mode in carla. If no camera sensor is needed, switching this on may speed up the simulating.

• auto_pilot

  Whether to use auto-pilot mode to control hero vehicle.

• n_vehicles

  Num of background vehicles in the world. Will be spawned at a random position.

• n_pedestrians

  Num of background pedestrians in the world. Will be spawned at a random position.

• disable_two_wheels

  Whether to disable two-wheel vehicles in NPCs.

• col_threshold

  The threshold of collision sensor intensity.

• waypoint_num

  The waypoints num ahead vehicle in waypoint_list of observations.

• verbose

  Whether print verbosely when running.

Users may need to check Carla documents for detailed information about configs.

Planner configurations

Here we explain the ‘planner’ key in config.

The Planner in Carla simulator is modified from default Carla planners. A global planner in Carla Python API is used to generate route and road options from start to end location. The local planner is defined in DI-drive to add navigation into observation. It provides current waypoint location, next waypoint location, road option command and distance to final target. Planners may also take nearby walkers, vehicles and traffic lights into account. Currently BasicPlanner and BehaviorPlanner are supported in DI-drive. For more details, check the API docs.

Some of the default config setting is as following:

planner = dict(
    type='basic',
    min_distance=5.0,
    resolution=5.0,
    fps=10.
)

• type

  Type of local planner

• min_distance

  Distance to find waypoint ahead vehicle.

• resolution

  Distance between waypoints when tracking route.

Note

You can define your own local planner in the simulator. Simply add the planner class in PLANNER_DICT in simulators.

Sensor configurations

Here we explain the ‘obs’ and ‘aug’ key in config

Sensor config is used to set up sensors and get image observation in simulation. There are several types of sensors and images available to add into config list. You can add more than one sensor of same type. They are tagged by provided ‘name’ key in config dict.

DI-drive simulator supports several types of sensor in Carla, including ‘rgb’ camera, ‘depth’ camera, ‘segmentation’ camera, and ‘lidar’ sensors. We also add ‘bev’ image to paint a bird-eye view image for the hero actor. It contains following information: road lines, lane lines, traffic lights, other vehicles, pedestrians, hero vehicles, forward waypoints list. Each stored in one channel (traffic lights in three channels). The simulator can only have one bird-eye view image to draw and store in sensor data. Here are the default configs for these sensors:

camera = dict(
    size=[384, 160],
    fov=90,
    position=[2.0, 0.0, 1.4],
    rotation=[0, 0, 0],
)

lidar = dict(
    channels=1,
    range=2000,
    points_per_second=1000,
    rotation_frequency=10,
    upper_fov=-3,
    lower_fov=-3,
    position=[0, 0.0, 1.4],
    rotation=[0, -90, 0],
    draw=False,
)

birdview = dict(
    size=[320, 320],
    pixels_per_meter=5,
    pixels_ahead_vehicle=100,
)

To add these sensors in simulator, you can add elements into ‘obs’ config list and change their default value:

obs=[
    dict(
        name='front_rgb',
        type='rgb',
    ),
    dict(
        name='top_rgb',
        type='rgb',
        position=[-2.8, 0, 5.5],
        rotation=[-15, 0, 0],
    ),
    dict(
        name='birdview',
        type='bev',
    ),
]

Then you can have ‘front_rgb’, ‘top_rgb’, ‘birdview’ image added into sensor data.

Cameras may need to add augmentation. for example, there may be two cameras with different type and same position to add random augmentation. So we need to store an augmentation setting in ‘aug’ key, which specifies the random value range of the augmentation.

aug=dict(
    position_range=[0, 0, 0],
    rotation_range=[0, 0, 0],
)

Running status

DI-drive simulator can get and record some running status in each frame, including basic status, navigation, information about simulation. They are automatically checked in each frame. Simulator provides interfaces to get dictionary about these info, stored in each key.

Status

• speed (float): speed in km/h

• location (np.ndarray): 3D location vector

• forward_vector (np.ndarray): 2D forward vector

• acceleration (np.ndarray): 3D acceleration vector

• is_junction (bool): Whether is at junction in Carla map

• tl_state (int): Traffic light state

  • 0: Red

  • 1: Yellow

  • 2: Green

  • 3: Off

• tl_dis (float): Distance between vehicle and current road end

Navigation

• agent_state (int): Agent navigation state

  • -1: Void

  • 1: Navigating

  • 2: Blocked by vehicle

  • 3: Blocked by walker

  • 4: Blocked by red light

• command (int): Road option during navigation

  • -1: Void

  • 1: Turn left

  • 2: Turn right

  • 3: Go straight at junction

  • 4: Follow lane

  • 5: Change lane to left

  • 6: Change lane to right

• node (np.ndarray): 2D location of current waypoint

• node_forward (np.ndarray): 2D direction of current waypoint

• target (np.ndarray): 2D location of target waypoint(the next of node)

• target_forward (np.ndarray): 2D direction of target waypoint

• waypoint_list (np.ndarray): Array containing waypoints on the route ahead. Each element is a 4D vector, consisting of the position and forward vector in x-y coordinates

• speed_limit (float): Speed limit of current position. It may come from the limit in current road and computed by planner

Information

• tick (int): Ticks for time in current episode

• timestamp (float): Real timestamp run in current episode

• total_lights (int): Total traffic lights met before

• total_lights_ran (int): Total red traffic light ran before