read-sensor

Run the save command with the sensor id and an output directory, for example read_sensor.py save --id 123 --out ./capture --seconds 10. Cameras write PNG files, lidar writes .ply point clouds, and other sensors append rows to data.jsonl.

Use the show command with a sensor id to open a pygame window rendering camera images or a top-down lidar scatter. A display is required; on headless machines use the save command instead. The grid command tiles multiple sensors in one window.

Calling save_to_disk inside the sensor listen callback deadlocks the client, measured stalling after a few hundred frames. The fix is lock-step capture: the client takes the world clock, ticks once per frame, and encodes images from the main loop via a queue.

Yes, when the server runs with --ros2 the sensor publishes native DDS topics such as rt/carla/<name>/image. The ros-info command lists topics, QoS, and enabled-for-ros state without ROS 2 installed; subscribers need the matching FastDDS profile and use_sim_time.

Yes, reading sensor data is unchanged on 0.10.0, including new sensors like fisheye cameras and hss_lidar. Note that enable_for_ros moved from carla.Actor to carla.ServerSideSensor, and rt/carla/map no longer exists.

CARLA cameras open underexposed and the auto exposure ramps over several frames. The save command discards a configurable number of warm-up frames (default 15) so the capture starts after exposure has settled.