Added
Updated
CARLA pairs a permissively licensed code base with open digital assets, so research groups can build and share environments without licensing negotiations. Its flexible API exposes traffic generation, pedestrian behaviour, weather, sensors and map control, and its sensor suite provides ground-truth semantic segmentation and depth that is impractical to collect on a real vehicle.
A scalable multi-client architecture lets several clients drive different actors in one world, and a non-rendering mode speeds up planning and control experiments that do not need graphics. The surrounding ecosystem, including the ROS bridge, ScenarioRunner, SUMO and Chrono co-simulation and third-party tool integrations, extends the simulator without modifying its core.
CARLA was created in 2016 at the Computer Vision Center in Barcelona with support from Intel Labs, and was introduced in 2017 in the paper CARLA: An Open Urban Driving Simulator at the first Conference on Robot Learning. The project was subsequently joined by the Toyota Research Institute and other sponsors.
The simulator is now co-hosted by the Computer Vision Center and the Embodied AI Foundation, and its maintenance is funded through a consortium of industrial members. The Computer Vision Center and Intel Labs received a Spanish National Research Award for the public-private partnership that produced CARLA.
Simulation has become a standard part of autonomous-driving development because it generates training data and reproduces rare failures without field operations. Demand has moved toward higher-fidelity environments, scenario-based validation and benchmarks that compare driving stacks on shared tasks.
CARLA serves that demand as a free, openly governed platform, and its challenge leaderboards give research groups and industrial teams a common measure. Sponsorship by semiconductor and automotive research organisations sustains the project, while the arrival of neural rendering and reconstruction tooling raises the fidelity bar for every open simulator.
The CARLA simulator itself is free and open-source under the MIT licence, with binaries, container images and source builds available at no cost. Revenue around the project comes from paid ecosystem services rather than from the software.
The CARLA team, spanning the Computer Vision Center and the Embodied AI Foundation, offers consortium agreements, custom software development, premium support, consulting services and ad-hoc collaboration for organisations that need guaranteed maintenance or bespoke simulation capability.
CARLA is an open-source simulator from the Computer Vision Center built to support the development, training and validation of autonomous driving systems. It ships with open digital assets, including urban layouts, buildings and vehicles, that can be used freely alongside the source code and protocols.
The platform runs on Unreal Engine as a scalable server multi-client architecture with a Python and C++ API, configurable sensor suites covering lidar, cameras, depth and GPS, an ASAM OpenDRIVE map pipeline, the ScenarioRunner traffic-scenario engine and a ROS bridge.
