Camera

Camera.set_poses(positions, orientations)

Update camera poses for all environments.

Parameters

• positions (Sequence[numpy.ndarray]): Sequence of length num_envs. Each entry represents the target camera position (x, y, z) in local coordinates.

• orientations (Sequence[numpy.ndarray]): Sequence of length num_envs. Each entry represents the target camera orientation as a quaternion (w, x, y, z).

---

Camera.set_depth_noise(noise_mean: float = 0.0, noise_std: float = 0.0)

Configure Gaussian noise applied to depth measurements.

Parameters

• noise_mean (float):

  Mean of the depth noise in world units.

• noise_std (float):

  Standard deviation of the depth noise in world units.

---

Camera.set_depth_mode(mode: str = "planar")

Set the depth representation mode used during rendering.

Parameters

• mode (str):

  Depth encoding mode:

  • "planar": z-distance along the camera’s forward axis (default).

  • "euclidean": Euclidean distance along the ray.

---

Camera.get_depth_map() -> torch.Tensor

Return the cached depth map from the last render(device) call. Every valid pixel in the depth map is inside [near_clip, far_clip], where the invalid ones are all zeros.

Returns

• depth_map (torch.Tensor): A tensor of shape (num_envs, height, width) with dtype torch.float32 on the same device as used in the last render(device) call, where:

  • num_envs is the number of environments.

  • height, width are the camera’s height and width.

---

Camera.get_segmentation_map() -> torch.Tensor

Return the cached instance segmentation map from the last render(device) call.

Returns

• segmentation (torch.Tensor):

  A tensor of shape (num_envs, height, width) with dtype=torch.int32 on the same device as used in the last render(device) call, where each pixel stores an instance ID:

• 0: background.

• >= 1: valid object instance IDs.

---

Camera.get_point_cloud() -> torch.Tensor

Unproject the cached depth map into a 3D point cloud in local coordinates.

Returns

• points (torch.Tensor):

  A tensor of shape (num_envs, max_points, D) with dtype torch.float32, where:

  • max_points is the maximum number of valid points (i.e., with non-zero depth) across all environments. For environments that have fewer valid points than max_points, it is padded with zeros.

  • D is the dimension of vertex positions (= 3; they are XYZ positions).

---

Camera.name

Return the name provided at construction via add_camera().