Screw Theory: Twists, Wrenches, and the Instantaneous Motion of a Rigid Body

What screw theory is and where it is used

Screw theory represents the paired quantities of rigid-body mechanics — angular-and-linear velocity, or moment-and-force — as single six-dimensional objects called screws. A twist is the velocity screw; a wrench is the force screw. Both live on the geometry of a line in space (its screw axis) plus a scalar pitch h that couples the linear and angular parts.

It is the standard language of modern robotics because it treats revolute and prismatic joints uniformly and gives coordinate-free, geometric insight that Denavit–Hartenberg tables hide. Concretely it appears in:

• Manipulator kinematics — the product-of-exponentials formula and the spatial/body Jacobian, whose columns are the joint twists.
• Grasping and force control — contact wrenches and grasp-map analysis (form/force closure).
• Multibody dynamics — Featherstone's spatial-vector articulated-body algorithm runs on twists and wrenches.
• Mechanism theory — mobility (Grübler count refinements) and singularity analysis of parallel robots like the Stewart platform.

How it works: the Chasles–Mozzi screw and the twist

The governing insight is the Chasles–Mozzi theorem: any instantaneous rigid-body motion equals a rotation of angular velocity ω about a unique axis, plus a translation along that same axis. Rotation and slide together sweep a helix — a screw.

A twist collects this as V = (ω, v), where ω is the 3D angular velocity and v = −ω × q + h·ω is the linear velocity of the body point currently at the origin; q is any point on the screw axis and h is the pitch. Define the pitch as

• h = (ωᵀv) / (ωᵀω) — the ratio of translation to rotation, in metres per radian.

Pure rotation gives h = 0; pure translation gives h = ∞ with ω = 0. The axis and unit direction are the Plücker coordinates of a line. Equivalently the twist is the 4×4 matrix [V] = [[ [ω]×, v ], [0, 0]] in the Lie algebra se(3); exponentiating it, T = exp([V]·θ), produces the finite screw displacement in SE(3) — the basis of the product-of-exponentials (POE) forward-kinematics formula T = e^{[S₁]θ₁}···e^{[Sₙ]θₙ}·M introduced by Brockett in 1984.

Key quantities and a worked example

Consider a revolute joint whose axis is the z-line through the point q = (2, 0, 0) m, spinning at ω = 3 rad/s about +z, with zero pitch (a hinge). The unit screw axis is ŝ = (0, 0, 1). The linear part is v = −ŝ × q = −(0,0,1) × (2,0,0) = (0, −2, 0). So the unit twist is

• S = (0, 0, 1, 0, −2, 0), and the actual twist V = 3·S = (0, 0, 3, 0, −6, 0).

A point on the body instantaneously at the origin therefore moves at v = (0, −6, 0) m/s — 6 m/s, exactly ω·r = 3 × 2, tangent to its 2 m circle. That is the value of the twist: one object gives every point's velocity via v_P = v + ω × p.

For the dual, a wrench F = (m, f) with a pure force f = (10, 0, 0) N applied at q = (0, 1, 0) m has moment m = q × f = (0,1,0)×(10,0,0) = (0, 0, −10) N·m. The instantaneous power a wrench does on a twist is the reciprocal product P = Vᵀ·F (using the swapped pairing ωᵀm + vᵀf).

Using screw theory in practice

In a real manipulator you build kinematics screw-by-screw:

• Pick a reference (home) configuration and write M ∈ SE(3), the end-effector pose there.
• Write each joint's unit screw Sᵢ = (ωᵢ, vᵢ) in the fixed frame: for a revolute joint, ωᵢ is the axis direction and vᵢ = −ωᵢ × qᵢ; for a prismatic joint, ωᵢ = 0 and vᵢ is the slide direction.
• Assemble the POE product T(θ) = e^{[S₁]θ₁}···e^{[Sₙ]θₙ}·M — no per-link frames, no DH table.

The spatial Jacobian Jₛ(θ) then has columns that are the joint twists expressed in the current configuration, mapping joint rates θ̇ to the end-effector twist: V = Jₛ·θ̇. Rank loss of Jₛ flags a kinematic singularity; two joint screws becoming linearly dependent (e.g., three parallel or coplanar axes) is instantly visible geometrically. To change reference frames, transform a twist with the 6×6 adjoint map Ad_T. For statics, wrenches transform by Ad_Tᵀ, and the same Jacobian gives joint torques τ = Jₛᵀ·F.

Screw theory versus its close cousins

Several representations describe rigid-body motion; screw theory's edge is unification and geometry.

• vs. Denavit–Hartenberg (1955): DH assigns four parameters per link and tabulates them — compact but frame-dependent, error-prone at the base/tool, and it obscures the physical axes. POE needs only two frames and exposes each joint's screw directly.
• vs. Euler angles / homogeneous transforms: T ∈ SE(3) stores a finite pose; a twist is its instantaneous generator (the element of se(3) whose exponential is T). Euler angles suffer gimbal lock; screws do not.
• vs. dual quaternions: mathematically equivalent (the dual-quaternion group double-covers SE(3)); dual quaternions are more efficient for interpolation and blending, while twists/wrenches read more directly for Jacobians and force analysis.
• vs. plain vector mechanics: screw theory is the coordinate-free packaging of the same Newton–Euler physics, exposing invariants (pitch, axis, reciprocity) that scalar bookkeeping hides.

Trade-offs, failure modes, and significance

Reciprocity is screw theory's sharpest tool and its subtlest pitfall. A wrench F and twist V are reciprocal when their reciprocal product vanishes, Vᵀ·F = ωᵀm + vᵀf = 0 — meaning the force system does no work on that motion. Constraint wrenches from joints are reciprocal to the freedoms they permit; getting the pairing (the swap of the two 3-vectors) wrong silently corrupts every power and torque calculation.

Practical caveats:

• Pitch singularities: h → ∞ for pure translation (ω = 0) needs careful limits; naive division by |ω| blows up.
• Small-angle exp/log: the SE(3) exponential and its logarithm need Taylor fallbacks near θ = 0 to stay numerically stable.
• Kinematic singularities: real, not artefacts — the mechanism loses a degree of freedom, and the Jacobian's screw columns become dependent.

Its significance: screw theory, from Ball's 1876 treatise to Brockett's 1984 POE and today's Modern Robotics curriculum, is the compact, geometric, singularity-transparent framework underpinning manipulator control, grasp planning, and multibody simulation.

Screw entities and how twists compare with alternative rigid-body motion representations

Representation	Dimensions / form	What it encodes	Best for
Twist V = (ω, v)	6-vector (se(3)); 4×4 matrix [ω]×,v	Instantaneous velocity: angular ω + linear v of a point	Velocity, Jacobians, POE kinematics
Wrench F = (m, f)	6-vector (dual space)	Static force f + moment m on a body	Statics, grasp analysis, force control
Screw axis + pitch h	line (Plücker) + scalar h + magnitude	Geometry of one motion/force	Visualizing ISA, singularity insight
Homogeneous matrix T ∈ SE(3)	4×4 (R, p)	Finite configuration (pose)	Storing/composing full poses
Denavit–Hartenberg	4 params/joint	Link frames of a serial chain	Legacy tabular kinematics
Euler angles + vector	3 + 3 scalars	Orientation + position	Human-readable pose, has gimbal lock