Skip to content

Default spaces and winding

RigExecJoint, RigExecControl, and volume weight providers evaluate their declared default-space channels as operations in the shared graph. Editing these inputs dirties their consumers without replacing the rig's compiled rig.

The formulas below use USD's row-vector matrix convention: the leftmost transform applies first. R is the existing orthonormalized rest transform, Rp the nearest namespace provider's rest transform, and Dp the resolved parent:defaultSpace. A default offset O applies the six default:* scalar channels: XYZ rotations in degrees followed by translation.

computed default:space = O * R * inverse(Rp) * Dp
avars:defaultSpace     = default:space            (fallback)
posed:defaultSpace     = avars:defaultSpace       (fallback)
parent:defaultSpace   = parent's computeDefaultFrame (fallback)
parent:space          = parent's computePointFrame   (fallback)

An absent namespace parent contributes identity. A provider's public computeDefaultFrame returns the four landmarks of its effective default:space. The matrix properties above can also be requested as ordinary computed property taps or connected to another declared input. These values are computed in memory; reading the USD attribute directly still reads its authored value.

Each matrix property follows the same selection rule: its connected computed matrix is authoritative, including an identity matrix. Otherwise a non-identity authored matrix is authoritative. An identity authored matrix selects the computed fallback. This matches the posed:space convention. To explicitly select world identity instead of an inherited space, connect a matrix-valued attribute containing identity.

When posed:space has no authoritative connection or non-identity authored value, posing uses:

posed = A * posed:defaultSpace * inverse(parent:defaultSpace) * parent:space

A applies per-axis avar scale, avar rotations in avars:rotationOrder, avars:rspin about X, and translation. avars:unitScaleFactor multiplies all three translation avars before composition, converting them to local distance units. It does not change rotation, scale, or default/rest translation. Volume weights retain rigid placement and use their inputs:scale* shape channels.

avars:rotationSign multiplies avars:rx/ry/rz -- and avars:rspin, which is an X rotation -- by a per-axis +1 or -1 before they are composed. It declares a MIRRORED control: two limbs built as reflections of each other answer the same typed rotation by turning into each other rather than the same absolute way, which is the convention an animator means by "behaviour mirroring". Each component carries only a sign; a magnitude is discarded and anything that is not negative -- including zero, which would otherwise erase an axis -- selects +1. Because it multiplies the avar and never the frame, it is the identity at rest: declaring it cannot move the rest pose, the skin, or anything constrained to the control. A manipulator reads and writes the EFFECTIVE angle, so the control turns the way it was dragged and the signed value is what reaches the stage.

With the default-space channels at their schema defaults, the formula reduces to A * R * inverse(Rp) * parentPosed. Editing a default pose does not edit the rest frame. Solver-owned joint poses remain authoritative.

For a RigExecTwistDistribution, inputs:twistTurns adds signed revolutions to the principal twist extracted from its endpoints. Sample weight w receives an additional 360 * twistTurns * w degrees around the aim axis. Integer turns retain the endpoint orientation while describing winding between the endpoints; fractional values also rotate the end and can be animated or connected. The C++ builder exposes SetTwistTurns, and Python exposes set_twist_turns.