API reference

Generated from the docstrings of the twinrobo package. Start with the core objects; a simulator adapter wraps a CameraTwin and renders its input from a simulator camera.

Area	Main entry points
Core	CameraTwin (optics → sensor → ISP), CameraSpec, CatalogRegistry, CameraFrame, stereo modules
Optics	DeepLensOptics, PSFBank and PSFCache, DenseDepthRenderer, LensRays, the lens-ray renderer, distortion
Sensor and ISP	sensor and ISP stage interfaces, color conversion
MuJoCo	MujocoCameraTwin, render backends, camera mounts
Isaac Sim	IsaacCameraTwin, the Isaac lens scene
LIBERO and RoboCasa	CameraTwinLiberoEnv, robots, demo replay, RoboCasa episodes
Validation	image, geometry and optics metrics

A typical use:

from twinrobo import CameraTwin
from twinrobo.plugins.mujoco import MujocoCameraTwin, MujocoRenderer

twin = CameraTwin.from_catalog("stereolabs/zed-x/2.2mm")      # a real camera
cam = MujocoCameraTwin(twin, MujocoRenderer(model, data), camera="wrist")
frame = cam.get_frame()                                         # CameraFrame
frame.rgb                                                       # the camera's image, on the GPU

Tensors follow one convention everywhere: rgb is linear [B, 3, H, W], depth is metric z-depth [B, 1, H, W] in meters: an input (it sets the defocus), and an output only with force_depth=True, as ground truth (see twinrobo.frame).