isaac-camera

Create an RtxCamera prim, wrap it in a CameraSensor with a resolution and annotator list such as rgb, then call play and read frames with sensor.get_data. Alternatively, define a UsdGeom.Camera and attach Replicator annotators to a render product.

Compute focal length as fx times sensor width divided by image width, set horizontal aperture to the sensor width, and derive vertical aperture from fy to preserve aspect. Set aperture offset attributes from the principal point offsets cx and cy relative to image center.

The legacy Camera class is now a stub; the implementation moved to isaacsim.sensors.experimental.rtx. New code should use RtxCamera for authoring plus CameraSensor for runtime capture, following the official migration table for renamed methods like set_focal_lengths.

Yes, apply the OmniLensDistortionOpenCvFisheyeAPI schema to the camera prim and set attributes such as fx, fy, cx, cy, k1, and imageSize. OpenCV pinhole and LUT-based distortion schemas are also supported.

Use TiledCameraSensor, which batches many cameras into a single render call, making it suitable for multi-view reinforcement learning data collection. SingleViewDepthCameraSensor is available for stereo-style depth simulation.

USD expresses focal length and apertures in tenths of a scene unit, so a 35 mm lens is 35 on a centimeter-stage but 0.025 on a meter-stage. Check your stage units and use the new set_focal_lengths API instead of the legacy setter.