carla

Built on Unreal Engine, CARLA simulates detailed city environments with accurate vehicle physics, pedestrian behavior, traffic signals, weather systems, and sensor modalities enabling researchers to test perception, planning, and control algorithms in diverse, controllable scenarios without physical vehicle access.

The simulator's sensor suite replicates the equipment used in real autonomous vehicles: RGB cameras with configurable field of view and resolution, LiDAR with adjustable beam count and range, radar, depth cameras, semantic segmentation cameras, instance segmentation, IMU, and GNSS.

Each sensor produces data in formats compatible with common robotics and ML pipelines, and the sensor placement, parameters, and number can be configured programmatically per experiment. CARLA's traffic manager controls the behavior of surrounding vehicles with configurable aggressiveness and compliance parameters.

CARLA is developed by the Computer Vision Center in Barcelona, with support from Intel and Toyota Research Institute, and is open-source under the MIT license.

It supports the OpenDRIVE standard for road network definition, enabling testing on both included city maps and custom environments built in the Unreal Editor or imported from real-world map data.

CARLA has been the primary simulation environment for major autonomous driving research competitions and benchmarks, and is used by research labs and automotive companies globally as a standard platform for algorithm development and ablation studies.