Install CARLA 0.9.16

This guide covers installing CARLA 0.9.16 on Ubuntu 24.04 accessed via SSH.

Environment

Machine	Lenovo Legion (Ubuntu 24.04)
CARLA path	/opt/carla/0.9.16
Python	System Python 3.12 (no need to install 3.10)
Access	SSH

Step 1. Download CARLA 0.9.16

note

If CARLA is already installed at /opt/carla/0.9.16, skip this step.

mkdir -p ~/carla && cd ~/carla

# Main package (~7GB)
wget https://carla-releases.s3.us-east-005.backblazeb2.com/Linux/CARLA_0.9.16.tar.gz

# Additional maps (optional but recommended for Town07 etc.)
wget https://carla-releases.s3.us-east-005.backblazeb2.com/Linux/AdditionalMaps_0.9.16.tar.gz

# Create installation directory and extract
sudo mkdir -p /opt/carla/0.9.16
sudo tar -xzf CARLA_0.9.16.tar.gz -C /opt/carla/0.9.16

If you downloaded the additional maps, sync them manually:

sudo mkdir -p /opt/carla/tmp
sudo tar -xzf AdditionalMaps_0.9.16.tar.gz -C /opt/carla/tmp

sudo rsync -av /opt/carla/tmp/CarlaUE4/ /opt/carla/0.9.16/CarlaUE4/
sudo rsync -av /opt/carla/tmp/Engine/ /opt/carla/0.9.16/Engine/

Step 2. Python Virtual Environment

CARLA 0.9.16 provides wheels up to Python 3.12 (cp312). Ubuntu 24.04 ships Python 3.12 by default, but Ubuntu 26.04 ships Python 3.14, which has no matching wheel. If you are on Ubuntu 26.04, install Python 3.12 first via the deadsnakes PPA:

sudo add-apt-repository ppa:deadsnakes/ppa
sudo apt update
sudo apt install python3.12 python3.12-venv python3.12-dev

note

Ubuntu 24.04 ships with Python 3.12 by default, so this step can be skipped on 24.04.

Then create the virtual environment using Python 3.12 explicitly:

python3.12 -m venv ~/carla-env
source ~/carla-env/bin/activate

pip install /opt/carla/0.9.16/PythonAPI/carla/dist/carla-0.9.16-cp312-cp312-manylinux_2_31_x86_64.whl
pip install numpy pygame

If you are using a different Python version, pick the matching wheel from the PythonAPI/carla/dist/ directory:

Wheel File	Python
carla-0.9.16-cp310-cp310-manylinux_2_31_x86_64.whl	Python 3.10
carla-0.9.16-cp311-cp311-manylinux_2_31_x86_64.whl	Python 3.11
carla-0.9.16-cp312-cp312-manylinux_2_31_x86_64.whl	Python 3.12

Step 3. Launch CARLA

warning

X11 forwarding does not work with CARLA. CARLA uses Vulkan for rendering, which is incompatible with ssh -X forwarding. Running ./CarlaUE4.sh over SSH will crash with:

VK_ERROR_INITIALIZATION_FAILED
Segmentation fault (core dumped)

Recommended: Headless Mode

Run CARLA without a display window. The Python API and ROS 2 bridge work fully in this mode:

/opt/carla/0.9.16/CarlaUE4.sh -RenderOffScreen

Optional Launch Flags

Flag	Purpose
-RenderOffScreen	No GUI window (required over SSH)
-quality-level=Low	Reduce GPU load
-world-port=2000	Change RPC port (default: 2000)

Step 4. Verify Installation

With CARLA running in one terminal, open another terminal and run:

source ~/carla-env/bin/activate
python3 -c "import carla; client = carla.Client('localhost', 2000); print(client.get_server_version())"
# Expected output: 0.9.16

Step 5. Remote Visualization Options

Since ssh -X cannot forward the CARLA window, use one of these options if a GUI is needed:

Option	Notes
VNC (tigervnc)	Full remote desktop; run CARLA inside VNC session
NoMachine	Smoother than VNC, free for personal use
VirtualGL + TurboVNC	Best performance, more complex setup

tip

For most use cases (Python API, ROS 2 bridge, RViz), headless mode is sufficient. RViz itself does work over ssh -X.

Step 6. Running CARLA

Two approaches are available depending on your setup.

Option A: System-wide Wrapper (Headless / SSH)

Suitable for SSH access or machines with a single GPU. Creates a carla command available to all users:

sudo vim /usr/local/bin/carla

Paste the following:

#!/bin/bash
exec /opt/carla/0.9.16/CarlaUE4.sh -RenderOffScreen "$@"

Make it executable:

sudo chmod +x /usr/local/bin/carla

Now any user can run:

carla                      # runs with -RenderOffScreen by default
carla -quality-level=Low   # additional flags can still be passed

Option B: NVIDIA GPU (Local Display)

On machines with both an integrated GPU and a discrete NVIDIA GPU (e.g. Lenovo Legion), Vulkan may default to the iGPU. Add this alias to your ~/.bashrc to force CARLA to use the NVIDIA GPU and render with a display window:

alias carla='VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/nvidia_icd.json \
__NV_PRIME_RENDER_OFFLOAD=1 \
__NV_PRIME_RENDER_OFFLOAD_PROVIDER=NVIDIA-G0 \
__GLX_VENDOR_LIBRARY_NAME=nvidia \
__VK_LAYER_NV_optimus=NVIDIA_only \
/opt/carla/0.9.16/CarlaUE4.sh -ResX=640 -ResY=640'

Apply immediately:

source ~/.bashrc

Then run:

carla

note

This alias launches CARLA with a visible window at 640×640. Adjust -ResX and -ResY as needed.

Notes

• No bundled libstdc++.so.6 found in CarlaUE4/Binaries/Linux/ — this build uses the system library directly and is compatible with Ubuntu 24.04 out of the box.
• The manylinux_2_31 wheel tag requires glibc ≥ 2.31; Ubuntu 24.04 ships glibc 2.39