louiszengcn/carlaair

CARLA-Air is a simulation platform that combines two existing simulation systems, CARLA for realistic urban driving and AirSim for physics accurate drone flight, into a single program. The goal is to let researchers test scenarios where ground vehicles and flying drones need to interact in the same simulated city, something that used to require running two separate simulators and bridging them together, which added delay and complexity. Instead of connecting two programs over a network, CARLA-Air modifies CARLA's game mode to also host AirSim's drone physics inside the same process, so there is no bridge and no added latency between the two systems. The authors say only about thirty five lines of code were changed in the upstream CARLA source to make this work, and the coordinate systems of both simulators are aligned so a position reported by one exactly matches the same position in the other. The platform supports up to eighteen different sensor types, including regular color cameras, depth cameras, semantic and instance segmentation, LiDAR, radar, and GPS, all synchronized across both ground vehicles and drones. Existing CARLA and AirSim Python scripts, as well as ROS 2 nodes, are reported to run without any code changes, and the project includes prebuilt ROS 2 examples. In testing, a typical workload combining vehicles, a drone, and eight sensors ran at close to twenty frames per second, and the authors ran the simulator continuously for three hours with hundreds of spawn and despawn cycles without any crashes. The project provides prebuilt executables for both Ubuntu and Windows so most users do not need to compile anything themselves, though building from source is also documented. The license is a custom non-commercial license, so review it carefully before using this for anything beyond research or personal projects.