CARLA Simulator

This repository contains a modified version of CARLA 0.9.0. The Python client provided in this repo can be used either with the provided Unreal project or a binary release of CARLA 0.9.0, which will be available here soon.

Requirements

• Python2.7
• cmake 3.9.0

Using the Unreal project

The following instructions assume CARLA will be built from source. Detailed instructions for CARLA are available here.

1. Ensure Unreal 4.19 is installed and UE4_ROOT is set to the root of your Unreal 4.19 directory. At the moment you need to compile and build it from source.

# e.g.
export UE4_ROOT=~/UnrealEngine-4.19/

2. Run:

• ./Update.sh, to download all CARLA content

3. make launch # Compiles CARLA and launches Unreal Engine's Editor.
4. make package # (Optional) Compiles CARLA and creates a packaged version for distribution.
5. make PythonAPI # To be able to run python client code, but make sure you already ran the previous steps excluding 4.

Useful commands

• Launching the editor

cd Unreal/CarlaUE4
~/UnrealEngine_4.19/Engine/Binaries/Linux/UE4Editor "$PWD/CarlaUE4.uproject"

• Compiling some changes without doing a full rebuild

cd Unreal/CarlaUE4
make CarlaUE4Editor

Useful Scripts

These utility scripts are found in the scripts folder.

load_carla_editor.bash

• Loads the Carla editor project found carla-pysim. Use -h for help, and it requires UE4_ROOT to be already set via .bashrc. To load another level (aside from the default), put the level name as an argument, such as: bahs load_carla_editor.bash FlatEarth

python_dependencies_install.bash

• Installs relevant python dependencies to ensure that the python client scripts execute properly

set_clang_X_X.bash

• Modifies the binary symlinks to point to a specific clang version, denoted by X_X. This assumes that that version for clang is already installed using apt-get

Using the packaged simulator

All the information below is discussed in the official CARLA docs. This is just an overview summary of some useful info. When you run the client code, make sure the server is running. Launch CARLA simulator server in standalone mode with default setting:

./CarlaUE4.sh

You could also launch with a set window size:

./CarlaUE4.sh -windowed -ResX=500 -ResY=500

Run client in manual control. The flag -m displays the live 2D map of cars moving.

cd PythonClient
python manual_control.py -m

You can modify and run the following script to send custom control commands:

python basic_path_demo.py

If you want to load a map, other than the default map, you can do so by running:

./CarlaUE4.sh /Game/Maps/Town02 -windowed -ResX=500 -ResY=500

Note: When specifying the map, make sure its the first argument, as there is no flag associated with it.

Now load the custom maps (for these make sure you are using the provided carla binaries as opposed to the official carla binaries):

./CarlaUE4.sh /Game/Maps/BasicPlane -windowed -ResX=500 -ResY=500 -world-port=2001

./CarlaUE4.sh /Game/Maps/FlatEarth -windowed -ResX=500 -ResY=500 -world-port=2001

If you are using the custom maps above, before you run scripts such as manual_control.py, you may want to disable the random spawning of non_player agents such as vehicles and pedestrian from inside CarlaSettings().

settings = CarlaSettings()
settings.set(
    NumberOfVehicles=0,
    NumberOfPedestrians=0)

Otherwise you will get random agents getting dropped into the scene and crashing.

Keyboard Inputs

Key	Action
W	Throttle
S	Brake
A D	Steer
Q	Toggle reverse gear
Space	Hand-brake
P	Toggle autopilot
← → ↑ ↓	Move camera
PgUp PgDn	Zoom in and out
Mouse Wheel	Zoom in and out
Tab	Toggle on-board camera
F11	Toggle fullscreen
R	Restart level
G	Toggle HUD
C	Change weather/lighting
Enter	Jump
F	Use the force
Alt+F4	Quit

Control commands

This is the structure used to send the vehicle control to the server.

Key	Type	Description
steer	float	Steering angle between [-1.0, 1.0] (*)
throttle	float	Throttle input between [ 0.0, 1.0]
brake	float	Brake input between [ 0.0, 1.0]
hand_brake	bool	Whether the hand-brake is engaged
reverse	bool	Whether the vehicle is in reverse gear

Time-stamps

Key	Type	Units	Description
platform_timestamp	uint32	milliseconds	Time-stamp of the current frame, as given by the OS.
game_timestamp	uint32	milliseconds	In-game time-stamp, elapsed since the beginning of the current level.

Debugging

• Error connecting to the server, even when the server is running. Since this is a TCP connection, it could fail sometimes due to server timed-out, server is doing a reset etc. If this persists, you may just want to restart the server.

ERROR: (localhost:2000) failed to connect: [Errno 111] Connection refused

For this one, kill and re-run the client.

ERROR: (localhost:2000) failed to read data: timed out

In the case when these connection issues persist, try changing the port number.

1. On the server-side:

./CarlaUE4.sh -windowed -ResX=500 -ResY=500 -world-port=2001

2. On the client-side:

python client_example.py --port=2001

Video Capture

The script client_example.py has an example for this. You need to add the following code snippet:

# Iterate every frame in the episode.
for frame in range(0, frames_per_episode):

    # Read the data produced by the server this frame.
    measurement, sensor_data = client.read_data()

    for name, measurement in sensor_data.items():
        filename = args.out_filename_format.format(episode, name, frame)
        measurement.save_to_disk(filename)

This however makes the simulation very laggy. The frames are stored under PythonClient/_out/

Cameras and sensor

Four different sensors:

• Camera: Scene final
• Camera: Depth map
• Camera: Semantic segmentation
• Ray-cast based lidar

Map customization

1. Follow the instructions on How to build CARLA on Linux
2. Map customization
3. Once you have generated maps using the editor, these maps should be under

UnrealEngine_4.19/carla/Unreal/CarlaUE4/Content/Maps

Next, go to the directory UnrealEngine_4.19/carla/Unreal/CarlaUE4/Config. Open the file DefaultGame.ini and add the maps to cook:

+MapsToCook=(FilePath="/Game/Maps/MyAwesomeMap")

Note: This needs to be confirmed, but I think when you generate maps, to be able to run demos such as manual_control.py, need to keep PlayerStarts assets which can be copied over from the DefaultMap.

4. With all the maps in place, now you need to create a binary version of CARLA. The recommended way is to use the Package.sh script provided. Alternatively, open the CarlaUE4 project via Unreal, go to the menu File -> Package Project, and select your platform, linux in this case. Before you run the Package.sh make sure UE4_ROOT is in the environment as it looks at the environment variable UE4_ROOT to find the right version of Unreal Engine.

export UE4_ROOT=~/UnrealEngine_4.19

5. If all the above steps go well, you should have the packaged binaries under the directory:

UnrealEngine_4.19/carla/Dist/0.9.0-24-xxxxx/LinuxNoEditor