View sim buffers as ``torch.Tensor`` (no-op if already a tensor).

Convert tensor to cuRobo device for isolated device management.

Get a skill from the registry.

Move all tensors to device

Converts quaternion from one convention to another. The convention to convert TO is specified as an optional argument. If to == 'xyzw', then

Draws a single line between two points.

Extract the planner's active joint state from full simulation joint state. cuRobo typically plans over a subset of "active joints" (`self._pl

Build the world state.

Compose pose in robot root frame to world frame.

Check if an obstacle is enabled in the collision checker. Args: collision_checker: cuRobo collision checker instance obstacle_nam

Get the poses of specific links in the robot root frame. Returns all links if link_names is None.

Initialize the AutoSim pipeline.

Create a pipeline instance from the registry. Args: id: The registered pipeline ID. **cfg_overrides: Optional keyword arguments t

Reset the environment.

Execute one step of the skill. Args: state: The current state of the world. Returns: The output of the skill

Return EE pose in world frame as [x, y, z, qx, qy, qz, qw] via cuRobo FK.

Reset the skill.

Set the current target object for selective collision checking. Args: target_object: Name of the target object, or None to disabl

Build a joint position tensor in cuRobo's joint order from Isaac Lab state. Isaac Lab and cuRobo use different joint orderings. We look up each c

Check if grid position is within bounds (internal numpy representation)

Create a BBoxCache that covers all USD purposes. Collision-only meshes (e.g. ``*/Collisions/*`` under Isaac Lab assets) are typically authored

Parse a '+x'/'-z'/'+x+y' axis string into a normalized direction vector.

Refine cuRobo world collision state before planning. This method synchronizes all dynamic collision geometry: - Rigid object poses (i

Clears all debug drawings.

Create a skill instance from the registry, extra_cfg will overwrite the default value in the skill configuration.

Get the end-effector pose of the robot.

Plan the skill. Args: state: The current state of the world. goal: The goal of the skill. Returns:

Plan a trajectory to one of K candidate target poses from a shared start joint state. Always uses cuRobo's batch planning under the

Register an apply method for a specific skill.

Update `pipeline._env_extra_info.object_reach_target_poses` from JSON.

modify the extra configuration of the skills.

Collect all primitive names from cuRobo world model that start with given prefix. Args: prefix: Prefix to filter primitive names

Compute reach goal by transforming object-frame offsets into robot root frame. Args: env: The Isaac Lab environment.

Project cached or current primary-frame offsets onto the target primary pose.

Compute the current rigid offset from the primary link frame to each extra link.

Use USD transform syncing instead of PhysX Fabric for Grasp Editor. Grasp Editor reads and displays ordinary USD prim transforms. With Fabric ena

Draw EE frame axes using debug lines (R=X, G=Y, B=Z).

Draw collision spheres as three orthogonal wireframe circles.

XY convex hull of world-frame points; requires at least 3 distinct vertices.

Get current primary and extra link poses in robot root frame.

Return (positions, radii) for all active collision spheres in world frame.

World transform: T_world_prim = T_world_body(scene) @ T_body_prim(usd_fixed).

Rasterize prims; return count successfully marked occupied. ``world_mat_for_path`` maps prim path → override world matrix (scene group). Missing

Transform Nx3 local points to world frame (USD row-vector convention).

Return ``(min_xyz, max_xyz)`` of a prim's world-space AABB, or None if it is empty/invalid. USD returns an empty ``Gf.Range3d`` (``min > max``, c

Convert world coordinates to grid coordinates (row, col) Args: pos: [x, y] in world frame Returns: grid_pos:

Apply the skill output to the environment. Args: skill: The skill instance. skill_output: The output of the

Extract the goal from the skill information. Args: skill_info: The skill information. env: The environment.

Get the configuration of the skill.

List all registered skill names.

Register a skill in the registry.

Check if the skill should be skipped for applying the action. Args: skill: The skill instance.

Visualize the debug target pose.

Visualize all reach target poses from env_extra_info as frame markers. Creates markers for: - ``env_extra_info.object_reach_target_poses`` un