CARLA

// Overview

CARLA is, in the project's own words, "an open-source simulator for autonomous driving research", developed from the ground up to support the development, training and validation of autonomous driving systems. Where a general game engine gives you a world, CARLA gives you a world plus the research scaffolding around it: open digital assets — urban layouts, buildings, vehicles — together with the protocols to drive them programmatically.

Two capabilities do most of the work for machine-learning teams. Sensor suites are specified flexibly, so a vehicle can be fitted with whatever combination of cameras, lidar and other sensors an experiment calls for, and the same scene re-run with a different rig. Environmental conditions are configurable in the same way, which is how a policy gets tested against weather and lighting it would take months to encounter on a real road. Together they make it practical to generate labelled training data and to replay the same scenario until a controller stops failing it.

The project is long-running and actively developed — the `ue5-dev` branch tracks the Unreal Engine 5 line — with documentation at carla-ue5.readthedocs.io covering building on Linux and Windows, a first-steps tutorial, the asset catalogue, the Python API reference and the blueprint library. Licensing is split the way simulators usually are: the CARLA code is MIT, the assets are CC-BY, and bundled dependencies such as PROJ, SQLite, Xerces-C and Eigen keep their own open-source licences.

// What it does

• Open digital assets: urban layouts, buildings, vehicles and pedestrians for driving research
• Flexible sensor-suite specification, so the same scene can be re-run with a different sensor rig
• Configurable environmental and weather conditions for stress-testing a policy
• Python API for scripting scenarios, collecting data and driving vehicles programmatically
• Unreal Engine 5 line under active development on the `ue5-dev` branch
• Documentation covering builds, first steps, the asset catalogue, the Python API and the blueprint library

// Getting started

CARLA is built from source against Unreal Engine. The commands below are the ones given in the `ue5-dev` README; the full build guides live at carla-ue5.readthedocs.io.

git clone -b ue5-dev https://github.com/carla-simulator/carla.git CarlaUE5

cd CarlaUE5
./CarlaSetup.sh --interactive

cmake --build Build --target carla-python-api-install

https://carla-ue5.readthedocs.io

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Reach for it when a driving policy has to be validated against scenarios that are unsafe or impractical to stage on a real road
• Reach for it when you need labelled perception training data with a sensor rig you control
• Reach for it when the same scenario must be replayed deterministically across model revisions
• Reach for it when testing how a controller degrades under weather and lighting changes

// How CARLA compares

CARLA alongside other open-source simulators & synthetic environments tools AI/TLDR tracks, ranked by GitHub stars.