carla-simulator/carla
Hardware & IoTOpen-source Unreal Engine driving simulator for developing, training and validating autonomous-driving systems.
Project overview
CARLA is an open-source urban driving simulator for autonomous-driving research. It provides the simulator code and protocols together with open digital assets such as urban layouts, buildings and vehicles, and it lets users specify sensor suites and environmental conditions flexibly. It is aimed at researchers and engineers who build, train and validate self-driving stacks, including perception, planning and control experiments.
The repository is the development branch for the Unreal Engine 5.8 version (ue58-dev), which exists alongside the UE 4.26 version in ue4-dev. The two versions differ significantly, so check which one fits your work before building. The project also maintains an ecosystem of related repositories: a leaderboard for validating autonomous-driving stacks, Scenario_Runner, a ROS bridge, driving benchmarks, and reinforcement-learning and imitation-learning code.
Build and first run
Hardware requirements are substantial: an Intel i7/i9 9th-11th gen or AMD Ryzen 7/9, at least 32 GB RAM, an RTX 3070/3080/3090/4090/5090 or better with 16 GB+ VRAM, and Ubuntu 22.04 or 24.04 or Windows 11. Ubuntu 20.04, Windows 10 and earlier are not supported.
Building requires access to the CARLA fork of Unreal Engine 5.8. Link your GitHub account to Epic Games first and keep your Git credentials available for the Unreal repository download.
-
Clone the branch:
git clone -b ue58-dev https://github.com/carla-simulator/carla.git CarlaUE5 -
On Linux, run the interactive setup from the repository root. It asks for your sudo password to install prerequisites and for GitHub credentials to fetch Unreal Engine:
cd CarlaUE5 ./CarlaSetup.sh --interactiveFor an unattended Linux install, export
GIT_LOCAL_CREDENTIALS=username@github_tokenand runsudo -E ./CarlaSetup.sh. On Windows, runCarlaSetup.bat; unattended mode is not available there. -
After setup, configure and build with CMake. On Linux:
cmake -G Ninja -S . -B Build --toolchain=$PWD/CMake/Toolchain.cmake -DCMAKE_BUILD_TYPE=Release -DENABLE_ROS2=ON cmake --build BuildOn Windows, run the configure command from the x64 Native Tools Command Prompt for Visual Studio 2022, omitting
-DENABLE_ROS2=ON. -
Install the Python API and launch the editor:
cmake --build Build --target carla-python-api-install cmake --build Build --target launch
The first-steps guide covers what to do after the simulator is running. Prebuilt releases are listed on the releases page, and the Python API reference documents client control of the simulation.
Constraints and licensing
Expect the build to take a long time and a significant amount of disk space, since it downloads and installs Unreal Engine 5.8 along with the prerequisites. To target an existing Python 3 installation, pass --python-root to the setup script and set -DPython_ROOT_DIR and -DPython3_ROOT_DIR in the CMake configure step. Unreal Engine 5 has its own license terms, separate from CARLA's.
CARLA-specific code is MIT-licensed, while CARLA-specific assets are CC-BY. Bundled dependencies carry their own licenses, including PROJ, SQLite, Xerces-C, Eigen, and the optional Autodesk FBX SDK.
HotRepo's take
The repository is a long-running research platform rather than a quick-start demo: the build footprint, Unreal Engine account requirement and GPU floor define who can realistically use it. On 2026-11-18 it recorded 27 new stars in the C++ daily list and reached 5th place on that list, a useful sign of continued interest in the UE5 branch even though no all-language daily rank is recorded. The surrounding benchmark, ROS and learning repositories matter as much as the simulator itself when evaluating whether it fits an existing self-driving workflow.
Written by AI from the project's README, updated . The project's own docs are the source of truth.