carla: an open-source urban driving simulator for training and validating autonomous driving systems

What it solves

CARLA provides a high-fidelity, open-source simulation environment for autonomous driving research. It eliminates the need for expensive and risky real-world testing by allowing developers to train, validate, and test autonomous driving systems in a safe, digital urban setting.

How it works

Built on Unreal Engine, CARLA simulates urban layouts, buildings, and vehicles. It allows researchers to specify flexible sensor suites and environmental conditions to mimic real-world driving. The platform provides a Python API for controlling the simulation and integrates with an ecosystem of tools for scenario execution, ROS connectivity, and benchmarking.

Who it’s for

It is designed for researchers and developers working on autonomous driving stacks, specifically those focusing on training and validating AI models for vehicle control and perception.

Highlights

• Open-source code and protocols with free digital assets like urban layouts and vehicles.
• Support for flexible sensor configurations and environmental conditions.
• Integration with a wide ecosystem, including a leaderboard for validation and bridges for ROS and AutoWare.
• Support for various AI training methods, including Conditional Imitation Learning and Reinforcement Learning.