nesl / ComplexEventSimulator

Complex Event Simulator

Complex Event Simulator

Complex Event Simulator

Setup

1. Run git clone --recurse-submodules https://github.com/nesl/ComplexEventSimulator.git
2. Install CARLA 0.9.10 with the Additional Maps. Extract the CARLA files and put them under a new directory with the name of CARLA_0.9.10.
3. Create virtual environment using the requirements.txt file.
4. Install Poetry into your system
5. Install Scenic by getting into the Scenic folder and using poetry install -E dev
6. Clone CARLA-2DBBox and name it bbox_annotation.

Run

1. Execute in one command line tab ./CARLA_0.9.10/CarlaUE4.sh
2. Use the next command to generate the scenario scenic pedestrian.scenic --simulate --param num_scenario 0 --param num_extra_pedestrians 0 --param output_dir camera_img --param bind_address '127.0.0.1' --param cameras_on "1,2"
   • num_scenario: choose one of the predefined scenarios encoded in the file
   • num_extra_pedestrians: how many extra pedestrians to render, besides the ones needed for the scenario. If no extra pedestrians put 0
   • output_dir: where to save images if a camera is instantiated
   • bind_address: which address to bind in order to allow for transmitting of camera frames, if you don't want to bind to anything just leave it as ""
   • cameras_on: select the cameras to instantiate. These cameras are defined in locations.txt. If you don't want to use any camera just leave it as ""

You can use the get_spectator_pos.py to check the location of any place in the simulation.

Changes to Scenic

• Added the capability to specify both depth, lidar and RGB cameras, as shown in the code snippet below. Because we want the bounding boxes of any objects we are interesed in, we create an RGB Camera coupled with a depth camera to do this, as needed by the functions used by the CARLA-2DBBox library. We specify the locations of these cameras, with c being a vector of the form [x,y,z,pitch,yaw,roll,camera id], and also define a behavior that takes the path where images and the bounding box information are going to be saved, and will make use of the actors_bb list with all the objects one wants to track.

behavior CameraBehavior(path):
    while True:
        take GetBoundingBox(actors_bb,path)


depth_camera = depthCamera at c[0] @ -c[1],
    with elevation c[2],
    with pitch c[3],
    with yaw c[4],
    with roll 0

rgbCamera at depth_camera,
    with elevation c[2],
    with pitch c[3],
    with yaw c[4],
    with roll 0,
    with depth depth_camera, 
    with camera_id int(c[-1][-1]),
    with behavior CameraBehavior(output_dir)

• The SetWalkDestination action is implemented in order to control where do pedestrians go. You can use it as follows: take SetWalkDestination(destination). Check the WalkAround behavior for more information on how to use it.

• And Object_Dummy auxiliary class was created in order to instantiate or eliminate any object of type static.prop (all objects that are not pedestrians or cars) at any point of the simulation. A code snippet to instantiate a box is shown below, an Object_Dummy object is initialized by specifying its blueprint, position, heading, and elevation.

obj_dumm = Object_Dummy(Uniform(*blueprints.boxModels),box_destination,0,0)
simulation().createObjectInSimulator(obj_dumm)

 To destroy the object:

obj_dumm.carlaActor.destroy()

• All static.prop objects have been associated with a static bounding box in order to work with the CARLA-2DBBox library.

• Control over vehicle's trajectories can be obtained using the BasicAgent class that provides CARLA. An example is provided with the CarFoBehavior behavior.

• To run any scenario use run_scenario(num_scenario=3,num_extra_pedestrians=100, output_dir="camera_img/")