CARLA is an open-source simulator designed for autonomous driving research, developed to support the development, training, and validation of autonomous urban driving systems. It provides open-source code, protocols, and digital assets (urban layouts, buildings, vehicles) that can be freely used. The simulator supports flexible sensor suite configurations and environmental conditions, making it suitable for studying various autonomous driving approaches. The paper evaluates three methods: a classic modular pipeline, an end-to-end model trained via imitation learning, and an end-to-end model trained via reinforcement learning. These methods are tested on controlled scenarios of increasing difficulty, and their performance is assessed using metrics provided by CARLA. The results highlight the importance of generalization and the challenges posed by complex urban environments. CARLA is designed to democratize research in autonomous driving and facilitate system verification, particularly in scenarios that are too dangerous or costly to test in the physical world.CARLA is an open-source simulator designed for autonomous driving research, developed to support the development, training, and validation of autonomous urban driving systems. It provides open-source code, protocols, and digital assets (urban layouts, buildings, vehicles) that can be freely used. The simulator supports flexible sensor suite configurations and environmental conditions, making it suitable for studying various autonomous driving approaches. The paper evaluates three methods: a classic modular pipeline, an end-to-end model trained via imitation learning, and an end-to-end model trained via reinforcement learning. These methods are tested on controlled scenarios of increasing difficulty, and their performance is assessed using metrics provided by CARLA. The results highlight the importance of generalization and the challenges posed by complex urban environments. CARLA is designed to democratize research in autonomous driving and facilitate system verification, particularly in scenarios that are too dangerous or costly to test in the physical world.