CARLA (Car Learning to Act) is an open-source simulator developed to support the training, testing, and validation of autonomous driving systems. Designed by the Computer Vision Center (CVC) and backed by contributions from the research community, CARLA is tailored for AI-based self-driving research with a strong focus on safety-critical scenarios and high-fidelity sensor simulation.

The platform is widely used by academic labs, research institutions, and R&D teams in the automotive and robotics industries. It provides a realistic urban driving environment, customizable traffic behaviors, and rich multimodal sensor outputs—including camera, LiDAR, radar, and GPS.

Key benefits of CARLA include:

• Simulation of complex driving scenarios for safer model development.

• Open-source and extensible framework, ideal for academic and industrial experimentation.

• High-fidelity sensor modeling for perception, planning, and control systems.

What are the main features of CARLA?

High-fidelity simulation environment

CARLA simulates realistic urban and suburban environments using the Unreal Engine, offering detailed visuals, physics, and object interactions.

• Weather, lighting, and time-of-day variations

• Configurable road networks and intersections

• Realistic vehicle dynamics and pedestrian behavior

Multimodal sensor support

The simulator provides a wide range of virtual sensors that mimic real-world hardware, allowing teams to test perception algorithms under controlled conditions.

• RGB, depth, and semantic cameras

• LiDAR and radar sensors

• GNSS, IMU, and ultrasonic sensors

• Configurable sensor noise and placement

Scenario-based testing

CARLA includes tools to design and execute complex driving scenarios involving traffic participants, dynamic obstacles, and scripted events.

• Trigger-based control of agents (vehicles, pedestrians)

• Scenario runner for automated test execution

• Support for rare edge cases and corner situations

API control and modular design

CARLA is built for integration with external AI models, providing APIs in Python and C++ for low-level control and data access.

• Interface with planning and control stacks

• Integration with reinforcement learning and imitation learning frameworks

• Modular architecture for sensor plug-ins and custom logic

Open research ecosystem

As an open-source project, CARLA fosters collaboration and benchmarking in the autonomous driving community.

• Public datasets and pre-configured maps

• Community-shared scenarios and tools

• Participation in international challenges like the CARLA Leaderboard

Why choose CARLA?

• Purpose-built for autonomous driving research, with realistic physics and customizable test environments

• Supports reproducible and safe experimentation, critical for evaluating AI models in edge cases

• Open-source and actively maintained, enabling full transparency and extensibility

• Rich sensor suite for testing perception systems, under a variety of simulated conditions

• Backed by an academic and industrial community, making it a standard in self-driving AI development