CARLA

Open-source simulator for autonomous driving research.

Get Started

Introduction

CARLA has been developed from the ground up to support development, training, and validation of autonomous driving systems. In addition to open-source code and protocols, CARLA provides open digital assets (urban layouts, buildings, vehicles) that were created for this purpose and can be used freely. The simulation platform supports flexible specification of sensor suites, environmental conditions, full control of all static and dynamic actors, maps generation and much more.

The current release, CARLA 1.0, runs on Unreal Engine 5.8. It makes CARLA an agentic platform, with state-of-the-art sensor simulation and all the maps back.

Get CARLA 1.0

Video

Highlighted features

  • Agentic CARLA: describe what you want and your AI coding agent does it. A library of 64 procedures vetted by the CARLA team covers setup, builds, packaging, asset import, sensors, scenarios, the Leaderboard and ROS 2, and the CARLA MCP server serves it to any agent that speaks the Model Context Protocol.
  • State-of-the-art sensor simulation: cameras (RGB, depth, segmentation, optical flow, fisheye), LIDARs, radar, GNSS, IMU and V2X. Ray-traced lens cameras are path traced through real lens models, and camera profiles are calibrated against real automotive datasets. Ground truth such as semantic and instance segmentation, depth and bounding boxes, is available for every frame.
  • NVIDIA ecosystem: NuRec turns recorded real-world drives into scenes CARLA can simulate, Cosmos world models turn CARLA drives into photoreal variations, Alpamayo 2 Super drives in closed loop, and DLSS 4.5 speeds up the cameras.
  • Unreal Engine 5.8: World Partition and Large World Coordinates for maps kilometres long, Lumen global illumination, hardware ray tracing and the path tracer.
  • Open content: 15 maps, from the classic towns to the large maps Town12, Town13 and Town15, plus 26 vehicles, 37 pedestrians, 82 props and dynamic weather. See the catalogue.
  • Scalability via a server multi-client architecture: multiple clients in the same or in different nodes can control different actors, and secondary servers render sensors on other GPUs.
  • Flexible API: CARLA exposes a powerful Python and C++ API that allows users to control all aspects related to the simulation, including traffic generation, pedestrian behaviors, weathers, sensors, and much more. The Python API supports Python 3.10 to 3.15.
  • Fast simulation for planning and control: no-rendering mode disables rendering to offer a fast execution of traffic simulation and road behaviors for which graphics are not required.
  • Maps generation: users can create their own maps following the ASAM OpenDRIVE standard via tools like RoadRunner, convert OpenStreetMap data, and share maps, vehicles and props as content packs without rebuilding CARLA.
  • Traffic scenarios simulation: ScenarioRunner (Python scenarios and OpenSCENARIO), Scenic and the CARLA Leaderboard define, execute and evaluate traffic situations.
  • Native ROS 2 integration: the server publishes the sensors and receives the vehicle control over ROS 2, with no bridge process, and the middleware (FastDDS, CycloneDDS or Zenoh) is chosen at launch.
  • Autonomous driving stacks: Autoware runs out of the box over native ROS 2, and NVIDIA Alpamayo 2 Super drives CARLA maps or NuRec scenes in closed loop.

Recommended reading

CARLA's functionality is covered extensively in the documentation. Here are some highlights covering some of CARLA's most useful and requested features.

Core features
  • Quickstart: Getting started with CARLA is easy, this guide will show you how to install and run the simulator.
  • Agentic CARLA: Connect your AI coding agent to CARLA and let it build, run and extend the simulator for you.
  • Actors: CARLA's actors are entities that interact within the simulation like vehicles, pedestrians and traffic signals. Get to know them here.
  • Sensors: CARLA boasts an impressive array of models of real world sensors like cameras, LIDAR and RADAR. The simulator also gives access to privileged information such as ground truth semantic segmentation and depth information.
  • Traffic Manager: CARLA's Traffic Manager controls NPCs to challenge your autonomous driving agent.
  • ROS 2: CARLA's native ROS 2 interface connects the simulator to the Robot Operating System without a bridge.
  • NVIDIA in CARLA: NuRec, Cosmos, Alpamayo and DLSS, and the workflows that combine them with CARLA.
Tutorials
  • First steps in CARLA: New to CARLA? Need some help with your first steps? This tutorial will get you up and running.
  • Bounding boxes: Need ground truth bounding boxes for vehicles or map features? This tutorial shows you how to access them through CARLA's API.
  • Creating custom maps for CARLA: Create your own custom maps for CARLA, using CARLA's built in asset library or importing your own assets.
  • Creating custom vehicles for CARLA: Create your own custom vehicles for CARLA.
  • Recording your simulation: Your CARLA simulations can be recorded and replayed exactly, enabling you to repeat and compare results for different sensors or configurations.

Do you like the project? Star us on GitHub to support the project!

Paper

CARLA: An Open Urban Driving Simulator

BiBTex

@inproceedings{Dosovitskiy17,
  title = { {CARLA}: {An} Open Urban Driving Simulator},
  author = {Alexey Dosovitskiy and German Ros and Felipe Codevilla and Antonio Lopez and Vladlen Koltun},
  booktitle = {Proceedings of the 1st Annual Conference on Robot Learning},
  pages = {1--16},
  year = {2017}
}