Carla

CARLA is the workhorse of autonomous driving research labs worldwide, and for good reason. It offers the most realistic open-source simulation environment for testing perception, planning, and control algorithms. The recent transition to Unreal Engine 5.5 (v0.10.0) brings substantial visual fidelity and physics improvements, keeping it competitive with commercial simulators. We'd reach for CARLA when we need full control over every aspect of the simulation—sensor configurations, weather, traffic, and map generation. The server multi-client architecture is a standout, allowing distributed simulation across machines. However, the learning curve is steep. Setting up CARLA requires familiarity with Python, the Unreal Engine ecosystem, and the ROS bridge if you're integrating with robotics stacks. Documentation is thorough but assumes a technical audience. Compared to SVL Simulator (now discontinued) or the proprietary NVIDIA DRIVE Sim, CARLA wins on openness and community, but loses on out-of-box ease and HD map support. The sensor suite is broad (LIDAR, radar, cameras, depth, IMU, GNSS) but not as physically modeled as some commercial offerings—you'll need to add your own noise models. For commercial product deployment, you'll likely need to extend the sim with custom assets and integrations. CARLA's licensing is MIT, so it's free for both research and commercial use, but support is community-driven. If you're a researcher or robotics engineer running reproducible experiments, CARLA is indispensable. If you're a student or hobbyist looking for a quick demo, consider the CARLA Leaderboard or pre-built Docker images to lower barriers.