CARLA

CARLA is an open-source simulation platform designed for autonomous driving research and development. CARLA provides a realistic virtual environment where researchers and engineers can test, train, and validate self-driving systems without using physical vehicles or public roads. The simulator includes detailed urban environments such as cities, highways, vehicles, pedestrians, and traffic behavior that represent real-world driving scenarios.

CARLA supports the full autonomous driving workflow. The simulator can generate synthetic data from virtual sensors including RGB cameras, depth cameras, LiDAR, radar, and GNSS. This synthetic data can be used to train machine learning models, evaluate perception pipelines, and test planning or control algorithms in repeatable conditions. Controlled simulation conditions also make it easier to test edge cases and safety-critical situations that are difficult to reproduce on physical roads.

CARLA uses a client-server architecture. The simulation server handles physics, rendering, and world state, while external client applications connect through APIs to control scenarios and collect data. CARLA provides Python and C++ APIs, which makes integration straightforward for machine learning frameworks, robotics stacks, and custom autonomous driving tooling.

SwissGPU workstations provide dedicated NVIDIA GPUs running on bare-metal hardware, which makes the infrastructure suitable for CARLA simulation workloads. You can install CARLA on the workstation and run high-fidelity simulation environments, generate training datasets, and train autonomous driving models directly on the same machine. Because each workstation provides exclusive access to GPU and system resources, CARLA workloads can run with predictable performance and consistent runtime behavior.

Resources

SwissGPU