Vaibhav Shende Vaibhav Shende

Installing CARLA 0.9.16 on Ubuntu 22.04: The Definitive Guide

Step-by-step walkthrough to install CARLA simulator on Ubuntu 22.04, with Python API setup, GPU acceleration, and common pitfalls.

Robotics

Installing CARLA 0.9.16 on Ubuntu 22.04: The Definitive Guide

What is CARLA?

CARLA (Car Learning to Act) is an open-source simulator for autonomous driving research. It provides:

  • High-fidelity 3D environments — realistic urban scenes with traffic, weather, and lighting
  • Python API — programmatic control of vehicles, sensors, and environment
  • Sensor simulation — cameras, LiDAR, radar, GNSS
  • Traffic simulation — realistic NPC behavior and interactions
  • Reproducible experiments — deterministic simulation for research

This guide covers the binary installation (recommended for most users). If you need to build from source, see the CARLA GitHub repository.

CARLA First View - Urban Environment CARLA 0.9.16 rendering a realistic urban environment with traffic and dynamic weather.

System Requirements

Minimum:

  • Ubuntu 22.04 LTS
  • 8 GB RAM
  • 20 GB free disk space
  • GPU: NVIDIA (GeForce GTX 1080 or better)

Recommended:

  • 16+ GB RAM
  • 50+ GB SSD space
  • GPU: RTX 2080 Super or better
  • CPU: 6+ cores

Note: CARLA can run on CPU, but GPU acceleration is strongly recommended for real-time simulation. CPU-only consume significant resources.

Step 1: Install NVIDIA Driver and CUDA Toolkit

CARLA requires NVIDIA drivers and CUDA for GPU acceleration. This is a critical step that often trips people up.

Detailed guide: See How I Set Up NVIDIA Driver and CUDA on My RTX 5070 Ti: A Real Setup Guide for a comprehensive, step-by-step walkthrough. That guide covers driver selection, CUDA compatibility matrices, environment setup, and common pitfalls. Recommended reading before proceeding.

Step 2: Download and Install CARLA 0.9.16

2.1 Download Binary Release

CARLA binaries are large (~8 GB). Choose a location with sufficient space. Download from the official CARLA 0.9.16 release page.

# Create a directory for CARLA
mkdir -p ~/carla-sim
cd ~/carla-sim
 
# Download CARLA 0.9.16
wget https://github.com/carla-simulator/carla/releases/download/0.9.16/CARLA_0.9.16.tar.gz
 
# Download additional maps (optional but recommended)
wget https://github.com/carla-simulator/carla/releases/download/0.9.16/AdditionalMaps_0.9.16.tar.gz

Download size: CARLA_0.9.16.tar.gz ~8 GB, AdditionalMaps ~14 GB. Total time: 30-60 minutes depending on internet speed.

2.2 Extract and Verify

# Extract main CARLA archive
tar -xzf CARLA_0.9.16.tar.gz
 
# Extract additional maps (optional)
tar -xzf AdditionalMaps_0.9.16.tar.gz
 
# Verify extraction
ls -la
# You should see: CarlaUE4.sh, PythonAPI, Import/ (if maps extracted), etc.
 
# Clean up tar files to save space
rm CARLA_0.9.16.tar.gz AdditionalMaps_0.9.16.tar.gz

2.3 Install Additional Maps

If you extracted AdditionalMaps_0.9.16.tar.gz, import the maps:

# The additional maps are in the Import/ directory
# The CARLA server will automatically import them on first launch
 
cd ~/carla-sim
./CarlaUE4.sh  # First run imports the maps (takes 5-10 minutes)

Available maps after import:

  • Town01 - Town02 (default, included in main package)
  • Town03 - Town15 (from AdditionalMaps package)

2.4 Make Launch Script Executable

chmod +x ~/carla-sim/CarlaUE4.sh

Step 3: Setup Python Environment

Create a virtual environment and install dependencies:

# Navigate to CARLA directory
cd ~/carla-sim
 
# Create virtual environment
python3 -m venv carla-venv
 
# Activate virtual environment
source carla-venv/bin/activate
 
# Upgrade pip
pip install --upgrade pip
 
# Install required dependencies
pip install numpy Pillow pygame psutil

Virtual environment: Recommended to isolate CARLA dependencies from your system Python. All subsequent pip install commands should be run inside this environment (with source carla-venv/bin/activate).

Step 4: Setup Python API

The Python API allows you to control CARLA via scripts.

4.1 Install CARLA Python Package

# Make sure virtual environment is active
source ~/carla-sim/carla-venv/bin/activate
 
# Navigate to PythonAPI directory
cd ~/carla-sim/PythonAPI
 
# Build and install the egg package
pip install -e .
 
# Or if the above fails, use:
python setup.py develop

4.2 Test Python API Import

# Make sure virtual environment is active
source ~/carla-sim/carla-venv/bin/activate
 
python -c "import carla; print(f'CARLA {carla.__version__} loaded successfully')"

Expected output:

4.26.2-0+++UE4+Release-4.26 522 0
Disabling core dumps.

If you get an import error, verify:

  1. CUDA libraries are in LD_LIBRARY_PATH
  2. Python version is 3.7 or higher
  3. pip installed the package (check pip list | grep carla)

Step 5: Launch CARLA

5.1 Start the CARLA Server

Open a terminal and run:

cd ~/carla-sim
./CarlaUE4.sh -world-port=2000 -quality-level=Epic

Parameters explained:

  • -world-port=2000 — Use port 2000 (default is 2000)
  • -quality-level=Epic — High graphics quality (use Low or Medium for slower hardware)
  • -headless — Run without display (for remote servers)
  • -fps=30 — Set simulation FPS (default 20)
  • -resx=1280 -resy=720 — Window resolution

First launch takes 5-10 minutes (imports additional maps if extracted). You should see:

LogCarlaServer: Welcome to CARLA 0.9.16
LogCarlaServer: Server listening on port 2000

5.2 Test Connection in New Terminal

Keep the server running. Open a new terminal:

cd ~/carla-sim
 
# Activate virtual environment
source carla-venv/bin/activate
 
# Test connection
python -c "import carla; client = carla.Client('localhost', 2000); print(f'Connected to CARLA {client.get_server_version()}')"

Expected output:

Connected to CARLA 0.9.16

Step 6: Verify Installation with Example Script

Create a simple script to spawn a vehicle:

# test_carla.py
import carla
import time
 
def main():
    # Connect to CARLA
    client = carla.Client('localhost', 2000)
    client.set_timeout(10.0)
    
    # Get world
    world = client.get_world()
    
    # Get spawn points
    spawn_points = world.get_map().get_spawn_points()
    
    # Blueprint library
    bp_lib = world.get_blueprint_library()
    vehicle_bp = bp_lib.find('vehicle.tesla.model3')
    
    # Spawn vehicle
    vehicle = world.spawn_actor(vehicle_bp, spawn_points[0])
    print(f"Spawned vehicle at {spawn_points[0].location}")
    
    # Apply control (move forward)
    control = carla.VehicleControl()
    control.throttle = 0.5
    vehicle.apply_control(control)
    
    # Wait
    time.sleep(5)
    
    # Cleanup
    vehicle.destroy()
    print("Vehicle destroyed")
 
if __name__ == '__main__':
    main()

Run the test:

# Activate virtual environment
source ~/carla-sim/carla-venv/bin/activate
 
python test_carla.py

You should see the vehicle spawn in the CARLA window and move forward.

Vehicle Spawned in CARLA Successful vehicle spawn in CARLA simulator.

Step 7: Explore CARLA Examples

CARLA ships with examples. Start the server, then in a new terminal:

source ~/carla-sim/carla-venv/bin/activate
cd ~/carla-sim/PythonAPI/examples

Essential examples:

ExamplePurpose
python tutorial.pySpawn vehicle with camera + autopilot
python manual_control.pyInteractive keyboard control (press ‘h’ for help)
python generate_traffic.py --number 50Spawn 50 NPCs with traffic behavior
python dynamic_weather.pyControl weather conditions
python synchronous_mode.pyDeterministic mode for ML pipelines
python automatic_control.pyAutonomous navigation with path planning
python start_recording.py --file log.logRecord simulation data
python start_replaying.py --file log.logReplay recorded data

For help: python <script>.py --help

Interactive control example:

Manual Control Interface - Pygame Display manual_control.py provides an interactive interface with real-time HUD showing vehicle telemetry, camera feeds, and control information.

Create a convenient startup script:

# ~/.carla_env
#!/bin/bash
export CARLA_HOME=~/carla-sim
export PYTHONPATH=$CARLA_HOME/PythonAPI:$PYTHONPATH
export LD_LIBRARY_PATH=/usr/local/cuda-12.4/lib64:$LD_LIBRARY_PATH
 
cd $CARLA_HOME
source carla-venv/bin/activate

Source it before each session:

source ~/.carla_env
./CarlaUE4.sh -quality-level=Epic

Step 9: Troubleshooting

Issue 1: “ImportError: No module named ‘carla’”

Cause: Python API not installed or wrong Python version.

Solution:

# Verify Python version
python3 --version  # Should be 3.7+
 
# Reinstall CARLA Python API
cd ~/carla-sim/PythonAPI
pip3 install -e .
 
# Or build from source
python3 setup.py develop

Issue 2: “Failed to connect to 127.0.0.1:2000”

Cause: Server not running or port blocked.

Solution:

# Check if server is running
ps aux | grep CarlaUE4
 
# Check if port 2000 is listening
lsof -i :2000
 
# Restart server
pkill -f CarlaUE4
cd ~/carla-sim && ./CarlaUE4.sh -quality-level=Epic

Issue 3: Low FPS or Stuttering

Cause: GPU not being used or too many actors.

Solution:

# Use lower quality level
./CarlaUE4.sh -quality-level=Low
 
# Reduce resolution
./CarlaUE4.sh -quality-level=Epic -resx=1280 -resy=720
 
# Check GPU utilization
watch -n 1 nvidia-smi

Step 10: Resources and References

Official Documentation

Community and Support

Step 11: Next Steps

1. Explore CARLA Examples

cd ~/carla-sim/PythonAPI/examples
python3 spawn_npc.py  # Spawn NPCs
python3 add_sensors.py  # Add camera and LiDAR
python3 manual_control.py  # Manual vehicle control

2. Read Official Documentation

Start with the CARLA Python API Documentation to understand available classes and methods.

3. Run Autonomous Driving Research

# Minimal autonomous agent example
import carla
import time
 
client = carla.Client('localhost', 2000)
world = client.get_world()
 
# Spawn vehicle
spawn_point = world.get_map().get_spawn_points()[0]
vehicle = world.spawn_actor(world.get_blueprint_library().find('vehicle.tesla.model3'), spawn_point)
 
# Add camera
camera_bp = world.get_blueprint_library().find('sensor.camera.rgb')
camera_transform = carla.Transform(carla.Location(x=2.5, z=0.7))
camera = world.spawn_actor(camera_bp, camera_transform, attach_to=vehicle)
 
# Control loop
for _ in range(100):
    vehicle.apply_control(carla.VehicleControl(throttle=0.5))
    time.sleep(0.05)
 
vehicle.destroy()
camera.destroy()

Step 12: Key Takeaways

  1. Binary installation is easiest; build from source only if needed
  2. Python API is the primary interface for research
  3. GPU acceleration is critical for real-time simulation
  4. First launch takes minutes — this is normal

Last updated: November 2026 | Tested on Ubuntu 22.04 with NVIDIA RTX 5070Ti

Back to Blog