CARLA installation

This document serves as supplementary material for installing CARLA from source. While not a step-by-step guide, it details my personal experience during the installation process, highlighting common issues and their workarounds.

📘 Guide for building CARLA from source (required for development)

Prerequisites

System requirements:

• Ubuntu 18.04
• 130 GB disk space
• 8GB dedicated GPU (recommended)

My system:

• Ubuntu 22.04.4
• 611 GB disk space
• NVIDIA GeForce GTX 1080 Ti (11GB)
• AMD Ryzen 7 2700 Eight-Core Processor
• 32GB RAM

Software requirements

The software requirements installation process went smoothly.

To clone CARLA’s Unreal fork, I followed the process outlined at https://www.unrealengine.com/en-US/ue4-on-github.

The Unreal Engine 4.26 build was successful, taking approximately 1.5 hours to complete. Here’s a screenshot of the Unreal editor to confirm the installation:

It’s worth noting that compiling all the required shaders to create a new project from a template took considerable time. The process even stalled once, requiring a restart. On the second attempt, it ran smoothly and opened up within seconds.

Build CARLA

After cloning CARLA, I switched to the master branch. According to the documentation:

The master branch contains the current release of CARLA with the latest fixes and features.

In this case, it’s version 0.9.15.

Compile Python API client

While attempting to compile the Python API client, I encountered the following error:

Setup.sh: Extracting boost for Python 3.
Building B2 engine..

A C++11 capable compiler is required for building the B2 engine.
Toolset 'clang' does not appear to support C++11.

> clang++ -x c++ -std=c++11  check_cxx11.cpp
check_cxx11.cpp:14:10: fatal error: 'thread' file not found
#include <thread>
         ^~~~~~~~
1 error generated.

** Note, the C++11 capable compiler is _only_ required for building the B2
** engine. The B2 build system allows for using any C++ level and any other
** supported language and resource in your projects.

You can specify the toolset as the argument, i.e.:
    ./build.sh [options] gcc

Toolsets supported by this script are:
    acc, clang, como, gcc, intel-darwin, intel-linux, kcc, kylix, mipspro,
    pathscale, pgi, qcc, sun, sunpro, tru64cxx, vacpp

For any toolset you can override the path to the compiler with the '--cxx'
option. You can also use additional flags for the compiler with the
'--cxxflags' option.

A special toolset; cxx, is available which is used as a fallback when a more
specific toolset is not found and the cxx command is detected. The 'cxx'
toolset will use the '--cxx' and '--cxxflags' options, if present.

Options:
    --help                  Show this help message.
    --verbose               Show messages about what this script is doing.
    --debug                 Build b2 with debug information, and no
                            optimizations.
    --guess-toolset         Print the toolset we can detect for building.
    --cxx=CXX               The compiler exec to use instead of the detected
                            compiler exec.
    --cxxflags=CXXFLAGS     The compiler flags to use in addition to the
                            flags for the detected compiler.

Failed to build B2 build engine
make: *** [Util/BuildTools/Linux.mk:142: setup] Error 1

Apparently, this is a well-known issue (https://github.com/carla-simulator/carla/issues/6901). Installing the g++ 12 compiler and recompiling solved the problem.

sudo apt install g++-12
make PythonAPI

Remember to restart your terminal or source your .bashrc file after setting the UE4_ROOT variable and before compiling the API. Failing to do so may result in the following error:

CMake Error at /usr/share/cmake-3.22/Modules/CMakeDetermineCCompiler.cmake:49 (message):
  Could not find compiler set in environment variable CC:


  /Engine/Extras/ThirdPartyNotUE/SDKs/HostLinux/Linux_x64/v17_clang-10.0.1-centos7/x86_64-unknown-linux-gnu/bin/clang.
Call Stack (most recent call first):
  CMakeLists.txt:33 (project)

CMake Error: CMAKE_C_COMPILER not set, after EnableLanguage
CMake Error: CMAKE_CXX_COMPILER not set, after EnableLanguage
-- Configuring incomplete, errors occurred!
See also "/home/dpascualhe/repos/carla/Build/libosm2dr-build/CMakeFiles/CMakeOutput.log".
make: *** [Util/BuildTools/Linux.mk:157: osm2odr] Error 1

To install the CARLA client in a virtual environment, you can use the .whl file generated during the compilation process. In my case, the generated wheel file is located at:

PythonAPI/carla/dist/carla-0.9.15-cp310-cp310-linux_x86_64.whl

As you can see, it’s been generated for Python 3.10. While it works for me, I’m unsure why only this version was generated. Here’s an installation example:

dpascualhe@dpascualhe:~/repos/carla$ python3.10 -m venv ~/.venvs/carla
dpascualhe@dpascualhe:~/repos/carla$ source ~/.venvs/carla/bin/activate
(carla) dpascualhe@dpascualhe-optical:~/repos/carla$ pip install PythonAPI/carla/dist/carla-0.9.15-cp310-cp310-linux_x86_64.whl
Processing ./PythonAPI/carla/dist/carla-0.9.15-cp310-cp310-linux_x86_64.whl
Installing collected packages: carla
Successfully installed carla-0.9.15

Compile the server

Server compilation (make launch) is very time-consuming due to the extensive shader compilation required. The process crashed multiple times, requiring repeated attempts. Eventually, I succeeded in compiling and launching the Unreal Engine editor. Below is a screenshot displaying the Town10 HD scenario:

It’s important to note that even after the editor launches, shaders and mesh distance fields continue compiling for some time.

To verify that graphical acceleration was working correctly, I ran nvidia-smi. The UE4Editor process was consuming approximately 4-5GB of my GPU’s VRAM.

Start the simulation

Once the editor launched and all shaders and mesh distance fields were ready, I attempted to run the simulation. However, when I pressed the PLAY button, the editor crashed. I discovered that my swap partition was too small for CARLA (https://github.com/carla-simulator/carla/issues/3398). After increasing my swap size to 8GB, everything ran smoothly (this likely would have prevented some earlier issues as well). This requirement was surprising, given that I have 32GB of RAM.

With the simulation running, I tested the Python client examples suggested in the documentation. While installing the Python requirements, I encountered an issue with the requested numpy version. The installation failed with this error message:

Collecting numpy==1.18.4 (from -r requirements.txt (line 3))
  Using cached numpy-1.18.4.zip (5.4 MB)
  Installing build dependencies ... done
  Getting requirements to build wheel ... done
  Preparing metadata (pyproject.toml) ... error
  error: subprocess-exited-with-error

  × Preparing metadata (pyproject.toml) did not run successfully.
  │ exit code: 1
  ╰─> [24 lines of output]
      Running from numpy source directory.
      <string>:461: UserWarning: Unrecognized setuptools command, proceeding with generating Cython sources and expanding templates
      Traceback (most recent call last):
        File "/home/dpascualhe/.venvs/carla/lib/python3.10/site-packages/pip/_vendor/pyproject_hooks/_in_process/_in_process.py", line 353, in <module>
          main()
        File "/home/dpascualhe/.venvs/carla/lib/python3.10/site-packages/pip/_vendor/pyproject_hooks/_in_process/_in_process.py", line 335, in main
          json_out['return_val'] = hook(**hook_input['kwargs'])
        File "/home/dpascualhe/.venvs/carla/lib/python3.10/site-packages/pip/_vendor/pyproject_hooks/_in_process/_in_process.py", line 149, in prepare_metadata_for_build_wheel
          return hook(metadata_directory, config_settings)
        File "/tmp/pip-build-env-es85oo_2/overlay/lib/python3.10/site-packages/setuptools/build_meta.py", line 373, in prepare_metadata_for_build_wheel
          self.run_setup()
        File "/tmp/pip-build-env-es85oo_2/overlay/lib/python3.10/site-packages/setuptools/build_meta.py", line 503, in run_setup
          super().run_setup(setup_script=setup_script)
        File "/tmp/pip-build-env-es85oo_2/overlay/lib/python3.10/site-packages/setuptools/build_meta.py", line 318, in run_setup
          exec(code, locals())
        File "<string>", line 488, in <module>
        File "<string>", line 465, in setup_package
        File "/tmp/pip-install-75ajk5k2/numpy_bd9cc42d6dae4f8da4480cc76d55c510/numpy/distutils/core.py", line 26, in <module>
          from numpy.distutils.command import config, config_compiler, \
        File "/tmp/pip-install-75ajk5k2/numpy_bd9cc42d6dae4f8da4480cc76d55c510/numpy/distutils/command/config.py", line 20, in <module>
          from numpy.distutils.mingw32ccompiler import generate_manifest
        File "/tmp/pip-install-75ajk5k2/numpy_bd9cc42d6dae4f8da4480cc76d55c510/numpy/distutils/mingw32ccompiler.py", line 34, in <module>
          from distutils.msvccompiler import get_build_version as get_build_msvc_version
      ModuleNotFoundError: No module named 'distutils.msvccompiler'
      [end of output]

  note: This error originates from a subprocess, and is likely not a problem with pip.
error: metadata-generation-failed

× Encountered error while generating package metadata.
╰─> See above for output.

note: This is an issue with the package mentioned above, not pip.
hint: See above for details.

Apparently, recent changes related to setuptools are causing this issue (https://github.com/numpy/numpy/issues/27405). None of the suggested workarounds were successful for me, so I opted for a simple solution: I removed the exact version match from the requirements.txt file and ran pip install after activating my carla virtual environment. Here’s my modified requirements file:

future
numpy
pygame
matplotlib
open3d
Pillow

Both the generate_traffic.py and dynamic_weather.py scripts ran flawlessly. However, I’ll keep an eye on potential issues that might arise from the installed numpy version. Here’s a screenshot of a running simulation: