The Robot Frame System

Any machine configured in Viam comes with the frame system service: an internally managed and mostly static system for storing the “reference frame” of each component of a machine within a coordinate system configured by the user.

Visualization of a wheeled base configured with motors and a mounted camera in the frame system tab of the Viam app UI

The frame system is the basis for some of Viam’s other services, like motion and vision. It stores the required contextual information to use the position and orientation readings returned by some components.

Used with

Configuration

To enable the default frame for a given component on a machine, navigate to the Config tab of the machine’s page in the Viam app and click Components. With mode as Builder, click Add Frame on the component’s card and Save Config.

To adjust the frame from its default configuration, change the parameters as needed for your machine before saving.

Navigate to the Config tab on your machine’s page in the Viam app, select the Builder mode, scroll to a component’s panel, and click Add Frame:

add reference frame pane

Select a parent frame and fill in the coordinates for translation (mm) and orientation (deg, rad, or q), according to the position and orientation of your component in relation to the parent frame.

{
  "components": [
    {
      "name": "<your_component_name_1>",
      "type": "<your_component_type_1>",
      "model": "<your_component_model_1>",
      "attributes": { ... },
      "depends_on": [],
      "frame": {
        "parent": "<world>",
        "translation": {
          "y": <int>,
          "z": <int>,
          "x": <int>
        },
        "orientation": {
          "type": "<ov_degrees>",
          "value": {
            "x": <int>,
            "y": <int>,
            "z": <int>,
            "th": <int>
          }
        }
      }
    }
  ]
}

Configure the reference frame as follows:

ParameterInclusionRequired
ParentRequiredDefault: world. The name of the reference frame you want to act as the parent of this frame.
TranslationRequiredDefault: (0, 0, 0). The coordinates that the origin of this component’s reference frame has within its parent reference frame.
Units: mm.
OrientationRequiredDefault: (0, 0, 1), 0. The orientation vector that yields the axes of the component’s reference frame when applied as a rotation to the axes of the parent reference frame.
Types: Orientation Vector Degrees, Orientation Vector Radians, and Quaternion.
GeometryOptionalDefault: none. Collision geometries for defining bounds in the environment of the machine.
Types: Sphere, Box, and Capsule.

For more information about determining the appropriate values for these parameters, see these two examples:

Visualize the frame system

You can visualize how your machine is oriented in the frame system in the Viam app. Navigate to the Config tab on your machine’s page, select mode as Builder, and click on Frame System.

The Viam app shows you a 3D visualization of the spatial configuration of the reference frames of components configured on your machine:

Default frame system configuration grid visualization for a single component, shown in the Frame System Editor

This tab provides a simple interface for simultaneously viewing and editing the position, orientation, and geometries of a machine’s components in the frame system.

For example:

Consider a machine configured with a jetson board, wired to a webcam camera and a wheeled base with two motors driving its wheels.

No reference frame configuration has been specified, so on the Frame System subtab of the Config tab, the components are shown to all be located on the default world origin point as follows:

Example machine’s default frame configuration shown in the Frame System Editor. All components are stuck on top of each other

The distance on the floor from the wheeled base to the board and camera setup is 200 millimeters.

Add this value to "X" in the base’s reference frame Translation attribute, and the frame system readjusts to show the base’s translation:

Base translated 200mm forwards shown in the Frame System Editor

The distance from the board to the camera mounted overhead is 50 millimeters.

Add this value to "Z" in the camera’s reference frame Translation attribute, and the frame system readjusts to show the camera’s translation:

Camera translated 50 mm overhead shown in the Frame System Editor

Now the distance between these components is accurately reflected in the visualization. However, the camera doesn’t yet display as oriented towards the base.

Adjust the orientation vector to 0.5 degrees in "OX" in the camera’s reference frame Orientation attribute, and the frame system readjusts to show the camera’s orientation:

Camera oriented .5 degrees OX shown in the Frame System Editor

Now that the frame system is accurately configured with the machine’s spatial orientation, motion service methods that take in reference frame information can be utilized.

Display options

Click and drag on the Frame System visualization to view the display from different angles, and pinch to zoom in and out:

Click the video camera icon below and to the right of the Frame System button to switch beween the default Perspective Camera and the Orthographic Camera view:

Bounding geometries

To visualize a component’s spatial constraints, add Geometry properties by selecting a component and selecting a Geometry type in the Frame System subtab of the Config tab of a machine’s page on the Viam app.

By default, a Geometry is shown surrounding the origin point of a component:

Demo robot with default box bounds added to the wheeled base, shown in the Frame System Editor

You can adjust the Size and Translation of a Geometry to change these bounds. For example:

Demo robot with translated box bounds added to the wheeled base, shown in the Frame System Editor

Demo robot with default sphere bounds added to the wheeled base, shown in the Frame System Editor

Demo robot with default capsule bounds added to the wheeled base, shown in the Frame System Editor

How the frame system works

viam-server builds a tree of reference frames for your machine with the world as the root node and regenerates this tree following reconfiguration.

Access a topologically-sorted list of the generated reference frames in the machine’s logs at --debug level:

an example of a logged frame system

Consider the example of nested reference frame configuration where two dynamic components are attached: a robotic arm, A, attaches to a gantry, G, which in turn is fixed in place at a point on the World of a table.

The resulting tree of reference frames looks like:

reference frame tree

viam-server builds the connections in this tree by looking at the "frame" portion of each component in the machine’s configuration and defining two reference frames for each component:

  1. One with the name of the component, representing the actuator or final link in the component’s kinematic chain: like "A" as the end of an arm.
  2. Another representing the origin of the component, defined with the component’s name and the suffix "_origin".

Access the frame system

The Robot API supplies the following method to interact with the frame system:

Method NameDescription
TransformPoseTransform a pose measured in one reference frame to the same pose as it would have been measured in another.

TransformPose

Transform a given source pose from the reference frame to a new specified destination reference frame. For example, if a 3D camera observes a point in space you can use this method to calculate where that point is relative to another object.

Parameters:

  • query (PoseInFrame): The pose that should be transformed.
  • destination (str): The name of the reference frame to transform the given pose to.
  • additional_transforms (Optional[List[Transform]]): A list of additional transforms.

Returns:

  • PoseInFrame (PoseInFrame): Transformed pose in destination reference frame.

For more information, see the Python SDK Docs.

# Transform a pose into the frame of myArm
first_pose = Pose(x=0.0, y=0.0, z=0.0, o_x=0.0, o_y=0.0, o_z=1.0, theta=0.0)
first_pif = PoseInFrame(reference_frame="world", pose=first_pose)
transformed_pif = await robot.transform_pose(first_pif, "myArm")
print("Position: (x:", transformed_pif.pose.x,
      ", y:", transformed_pif.pose.y,
      ", z:", transformed_pif.pose.z, ")")
print("Orientation: (o_x:", transformed_pif.pose.o_x,
      ", o_y:", transformed_pif.pose.o_y,
      ", o_z:", transformed_pif.pose.o_z,
      ", theta:", transformed_pif.pose.theta, ")")

Parameters:

  • ctx (Context): A Context carries a deadline, a cancellation signal, and other values across API boundaries.
  • pose ([PoseInFrame]): The pose that should be transformed.
  • dst (string): The name of the reference frame to transform the given pose to.
  • additionalTransforms (Optional[LinkInFrame]): A list of additional transforms.

Returns:

For more information, see the Go SDK Docs.

// Define a Pose coincident with the world reference frame
firstPose := spatialmath.NewPoseFromPoint(r3.Vector{X: 0.0, Y: 0.0, Z: 0.0})

// Establish the world as the reference for firstPose
firstPoseInFrame := referenceframe.NewPoseInFrame(referenceframe.World, firstPose)

// Calculate firstPoseInFrame from the perspective of the origin frame of myArm
transformedPoseInFrame, err := robot.TransformPose(ctx, firstPoseInFrame, "myArm", nil)
fmt.Println("Transformed Position:", transformedPoseInFrame.Pose().Point())
fmt.Println("Transformed Orientation:", transformedPoseInFrame.Pose().Orientation())

Additional transforms

Additional transforms exist to help the frame system determine the location of and relationships between objects not initially known to the machine.

For example:

  • In our example of nested dynamic attachment, the arm can be managed by the frame system without additional transforms because the base of the arm is fixed with respect to the gantry’s platform, and the gantry’s origin is fixed with respect to the world reference frame (centered at (0, 0, 0) in the machine’s coordinate system).

    However, an arm with an attached camera might generate additional information about the poses of other objects with respect to references frames on the machine.

    With the vision service, the camera might detect objects that do not have a relationship to a world reference frame.

    If a camera is looking for an apple or an orange, the arm can be commanded to move to the detected fruit’s location by providing an additional transform that contains the detected pose with respect to the camera that performed the detection.

    The detecting component (camera) would be fixed with respect to the world reference frame, and would supply the position and orientation of the detected object.

    With this information, the frame system could perform the right calculations to express the pose of the object in the world reference frame.

Usage:

  • You can pass a detected object’s frame information to the supplemental_transforms parameter in your calls to Viam’s motion service’s GetPose method.
  • Functions of some services and components also take in a WorldState parameter, which includes a transforms property.
  • TransformPose has the option to take in these additional transforms.