Abstract

We introduce CARLA, an open-source simulator for autonomous driving research. CARLA has been developed from the ground up to support training, prototyping, and validation of autonomous urban driving models, including both perception and control. In addition to open-source code and protocols, CARLA provides open digital assets (urban layouts, buildings, vehicles, pedestrians, etc.) that were created specifically for this purpose and can be used and redistributed freely. The simulation platform supports flexible specification of sensor suites and a wide range of environmental conditions. Using the presented simulation platform and content, we study the performance of two approaches to autonomous urban driving: a classic modular rule-based pipeline and an end-to-end model trained via imitation learning. The approaches are evaluated on a series of controlled scenarios of increasing difficulty, and their performance is examined in detail via metrics provided by the platform, illustrating the platform’s utility for research on autonomous urban driving.

Materials