Foundations T26 / 27 · Contact and Convex Geometry
Contact Kinematics: Why Rolling Is Not «Sliding Along»
A finger rolls 30° over an object's surface — how far does the contact point travel on each side? A ball rolls around a square back to its start — how much has its orientation drifted? Rolling is governed by a set of first-order equations, and the circular path even has a closed-form solution.
One-line takeaway
The pure-rolling constraint locks the arc-length migration of the contact point on the two surfaces to be equal; a ball rolling one full lap along a circular path ends up rotated by 2π√(1+(ρ/R)²) about the axis (−ρ/R, 0, −1) — it returns to its starting position, but not to its starting orientation.
In-hand object rotation, regrasp planning, the «physical legality» of teleoperation data — all of it stands on the rolling equations. Why finger surfaces are designed with curvature, why fingertips are arc-shaped — the answers all live in this lesson's geometry.
Why this tool exists
A wine bottle tips over: did it roll here or slide here? Most people answer correctly (roll), but cannot state the criterion. The answer is the contact-point velocity: rolling = the contact point's instantaneous velocity is zero; sliding = it is not. This definition looks like mere «moving or not moving», but it is extremely powerful — it locks the migration paths of the contact point on the two bodies' surfaces together.
A finger rolls a short distance over an object's surface: how far does the contact point move on the finger surface, and how far on the object surface? The two numbers must be equal (arc-length conservation, proved below) — this constraint is exactly the criterion for «is the finger slipping or rolling» in tactile teleoperation: does the sensor see the tangential field translate as a whole (slip), or an asymmetric leading-edge/trailing-edge migration (roll)?
The most counterintuitive phenomenon: a ball rolls one lap around a square path back to its start and its orientation does not come back — with L/R = 1 it is off by 39.6° about the vertical. This is not numerical error; it is holonomy, the classic geometric phase of nonholonomic constraints. The orientation change after one lap of circular rolling has a closed-form solution, fully derived in this lesson.
Definitions
Definition 4.1 The pure-rolling constraint
Two rigid bodies in contact; the material points at the contact point on each body are called A (body 1) and B (body 2). Pure rolling = at every instant A and B have equal spatial velocities (zero relative velocity, and no relative sliding along the tangent).
Why define it this way«Equal contact-point velocities» is stronger than «no relative sliding»: it also locks the tangential velocity difference. Note that the contact point is not a fixed material point — as rolling proceeds, A and B each migrate over their own surface.
Definition 4.2 Angular velocity of plane–sphere rolling
The ball center moves with velocity v (horizontal), ball radius R, vertical unit vector ẑ. Under pure rolling (no spin) the ball's angular velocity is ω = (ẑ × v)/R.
Why define it this wayThis is the conclusion of Theorem 4.1, stated up front as working vocabulary. Meaning of the components: rolling toward +x, the ball rotates about −y (by the right-hand convention) — «the rolling axis is always perpendicular to the direction of motion, and horizontal».
Definition 4.3 Holonomy
The difference between the orientation after one lap along a closed path and the initial orientation. Holonomy is not error; it is an intrinsic phase of the constraint geometry: the path's shape (up to scaling) fully determines it.
Why define it this wayA finger wipes one lap around an object's surface; a teleoperator rubs one lap around the object — when they return to the start, the finger's orientation has drifted. Without understanding holonomy you will mis-book this as «control error».
Theorems and proofs
Theorem 4.1 Angular-velocity formula for plane–sphere rolling
A ball rolls on a plane without spinning, ball-center velocity v = (v_x, v_y, 0); then the angular velocity is ω = (ẑ × v)/R, i.e. ω = (−v_y, v_x, 0)/R.
Proof
Spatial position of the contact point (a material point on the ball): p_c = p_center − Rẑ. Its velocity = ball-center velocity + ω × (−Rẑ).
Pure rolling requires the contact-point velocity to be zero (the plane is stationary): v + ω × (−Rẑ) = 0.
Let ω = (ω_x, ω_y, ω_z). Compute ω × (−Rẑ) = −R(ω × ẑ), component by component: ω × ẑ = (ω_y·1 − 0, 0 − ω_x·1, 0) = (ω_y, −ω_x, 0). So the constraint gives v − R(ω_y, −ω_x, 0) = 0, i.e. v_x = Rω_y and v_y = −Rω_x.
Solving: ω_x = −v_y/R and ω_y = v_x/R; ω_z does not appear — the vertical spin component is unconstrained (the ball spins about the contact normal while the contact point stays put). The pure-rolling definition adds «no spin» to set ω_z = 0. Combining:
Uniqueness check: the constraint involves only the horizontal components of ω, and ω_z is pinned by the no-spin assumption; without that assumption the solution set is the affine line ω + cẑ. ∎
NoteThe spin degree of freedom is not a quirk — it is exactly what your fingers use to twist a bottle cap. Rolling contact naturally makes «rotation about the normal» a free degree of freedom: the least-effort channel for rotating objects in hand.
Theorem 4.2 Conservation of arc-length migration
Under pure rolling, over a small time interval dt the arc-length element swept by the contact point on body 1's surface equals the arc-length element swept on body 2's surface.
Proof
Over dt the contact point moves in space from p to p + dp (dp is tangent to both surfaces — contact is maintained). The contact point's migration on body 1: body 1 rotates by ω₁dt and translates by v₁dt; the change of the contact point's material coordinates on body 1 is determined by the relative velocity.
Key point: the spatial velocity of point A (the contact material point on body 1) = v₁ + ω₁ × r_A (r_A points from the body center to A). Likewise for B. Pure rolling: v_A = v_B = the common velocity ū.
A's material migration rate over body 1's surface = |v_A − v₁| (freeze body 1; A moves relative to it) = |ū − v₁|. Likewise B's on body 2 = |ū − v₂|.
For sphere–plane: v₁ = v (ball-center velocity), ū = 0 (the contact point is instantaneously at rest) ⟹ A's migration rate = |v| = |ū − v₂| = |0 − 0|? — wait: B is a material point on the plane; the plane is at rest, v₂ = 0, B's spatial velocity is zero — so B's migration rate on the plane is zero? Contradiction?
Resolving the puzzle: the plane is infinite, and «migration of the contact point over the plane» does exist — the contact point moves in space by dp = ū dt = 0? The contact point's spatial velocity is ū = 0 (by the rolling definition), but as a «spatial event» the contact point appears at a neighboring position in the next instant — because once the ball advances, a new lowest point occupies a different spatial position. The correct quantity: the contact point's spatial migration rate = |v| (ball-center velocity); it sweeps path length |v|dt on the plane and simultaneously sweeps arc length |ω|·R·dt = (|v|/R)·R·dt = |v|dt on the sphere. Equal. ∎
Rigorous statement: equality of the arc-length elements on the two surfaces follows from «continuity of the contact point's spatial position» + normal conformity of the two surfaces. For general surfaces this is arc-length conservation: ds₁ = ds₂.
NoteThat «near contradiction» inside the proof is worth savoring: the contact point is not a material point; it is an event in space. Confusing «material-point velocity» with «contact-point migration» is the most common mistake in this class of problems.
Theorem 4.3 Closed-form solution for rolling along a circular path
A ball (radius R) rolls one lap along a circle of radius ρ on a plane: the net orientation = rotation by 2π√(1+(ρ/R)²) about the axis (−ρ/R, 0, −1)/√(1+(ρ/R)²). It returns to its start, but its orientation does not return.
Proof
Rolling at constant speed along the circular path, the contact point's spatial position is p(t) = ρ(cos Ωt, sin Ωt, 0) and the ball-center velocity is v = ρΩ(−sin Ωt, cos Ωt, 0). By Theorem 4.1: ω = (ẑ×v)/R = (ρΩ/R)(−cos Ωt, −sin Ωt, 0).
Directly integrating the attitude equation Ṙ = [ω]× R is awkward (the direction of ω rotates). Switch to a frame that rotates with the path: write R = R_z(Ωt)·S, where S is the orientation in the «rotating with the contact point» frame.
Deriving S's equation: Ṙ = [Ωẑ]× R_z S + R_z Ṡ, while Ṙ = [ω]× R_z S. Left-multiplying by R_zᵀ gives Ṡ = [R_zᵀω − Ωẑ]× S.
Computing R_zᵀω: apply R_z(−Ωt) to (−cosΩt, −sinΩt, 0)·(ρΩ/R): after rotating by −Ωt, x' = −cos²Ωt − sin²Ωt = −1 and y' = cosΩt sinΩt − sinΩt cosΩt = 0. Hence R_zᵀω = (−ρΩ/R, 0, 0).
So Ṡ = [(−ρΩ/R, 0, 0) − Ω(0,0,1)]× S: a rotation driven by the constant angular velocity u = Ω·(−ρ/R, 0, −1). Solution: S(t) = exp([u]× t).
One lap T = 2π/Ω: the net orientation = R_z(ΩT)·S(T) = R_z(2π)·S = S (R_z(2π)=I). So the net orientation = rotation by |u|T = 2π√(1+(ρ/R)²) about the axis (−ρ/R, 0, −1) (normalized). ∎
NoteThree things worth pausing on: (1) the rotation angle is independent of how fast you roll (Ω cancels) — holonomy is a geometric quantity, not a kinematic one. (2) It depends only on the ratio ρ/R — scale invariance (numerical check: the square-path net orientations for L=2,R=1 and L=1,R=0.5 are elementwise equal). (3) Only the rotation angle 2π√(1+(ρ/R)²) mod 2π is the «visible» deviation: for ρ/R=0.5, 7.0248 rad ≡ 0.7416 rad, which is exactly what numerical integration gives. Max matrix-element error between closed form and numerical integration: 8.7×10⁻⁸.
Work it by hand
A cylindrical finger (radius 20 mm) pure-rolls 30° on a plane: how far does the contact point move on the plane? How far on the cylinder? A ball (R = 1 m) rolls one lap along a square of side 1 m — how much does it deviate about the vertical?
- Cylinder: the pure-rolling constraint (Theorem 4.2) ⟹ the arc lengths on the two sides are equal: s = Rθ = 20×(π/6) = 10.47 mm. Both sides are 10.47 mm — the equality is a constraint, not a coincidence.
- Ball: the numerical experiment gives, for L/R = 1, a z-angle of the net orientation = −39.62° (atan2 reading; the full orientation is a 3D rotation and the z-angle is only one projection of it). Shrinking the ball to R=0.5 and the square to 0.5 m on the same path: the orientation is elementwise identical (scale invariance).
- Comparison: rolling 1 m in a straight line gives z-angle = 0.0000° — straight-line rolling only rotates about a horizontal axis; no holonomy. Holonomy requires accumulated changes of the path's direction (curvature).
10.47 mm on each side; z-angle −39.62° (the net orientation is a 3D rotation about a tilted axis). The arc-length equality is «mundane» and the holonomy «surprising» — both follow from the pure-rolling constraint.
Run the numbers
Where the numbers come from
Numerically integrating ω = (ẑ×v)/R for the orientation (Rodrigues update, 6000 steps):
- Closed-form check on the circular path: max matrix-element error 8.7e-08 / 1.5e-07 / 1.0e-06 for ρ/R = 0.5/0.6/2.0
- Scale invariance: elementwise difference of the net orientation matrices for (L=2, R=1) vs (L=1, R=0.5) = 0.00e+00
- Straight-line comparison: z-angle after rolling 1 m = 0.0000° (rotation about a horizontal axis only, no holonomy)
- z-angles of the square for each L/R: 0.5→−13.11°, 1→−39.62°, 2→+101.62°, 4→+86.95° (non-monotonic — the net orientation is a 3D rotation; the z-angle is only a projection)
Where this shows up
| The object here | The counterpart in RL | Used in |
|---|---|---|
| Pure rolling constraint and contact-point migration | The criterion for «is the finger rolling or sliding» in teleoperation data: tangential field translating as a whole = slip; asymmetric migration = roll | used in → |
| The spin degree of freedom ω_z | The mechanical channel for twisting a bottle cap; the least-effort direction for in-hand rotation | used in → |
| Holonomy and scale invariance | Where «the finger comes back skewed after one lap» in regrasp planning comes from — not control error | used in → |
| Contact kinematics equations | The basis for finger-surface curvature design: an arc-shaped fingertip keeps contact migration controllable | used in → |
Boundaries: where this article stops
Where this lesson stops: Montana's contact kinematics equations for general surfaces (1988, how curvature tensors evolve with contact), rolling–sliding mixtures and transition criteria, spin friction, and deformable rolling within a soft fingertip's contact patch — none covered. Honest disclaimer: real fingertips have large-area soft contact, and rolling is partially replaced by local deformation — the rigid-finger theory gives trends and criteria; quantitative design needs a soft-contact model on top.
Self-test: 4 questions
All hand-computation and «find the flaw in this proof» questions — nothing to memorize. Stuck? Go back to the section.
01When a wine bottle tips over, does it roll or slide — what is the criterion?
Look at the contact point's instantaneous velocity: zero means rolling, nonzero means sliding. The ground's friction on the bottle tends to maintain rolling (static friction adapts), so on dry ground it usually rolls and on ice it easily slides. The criterion is kinematic (contact-point velocity), not «it looks like it's spinning».
02A finger rolls 10° over an object's surface: how much arc length does the contact point travel on the finger surface and on the object surface?
Equal: s = R_eff·θ, where R_eff is the effective radius of curvature (Theorem 4.2). Equal arc-length migration on both sides is exactly the physical basis for distinguishing «roll» from «slip» by touch — in slip the tangential field translates as a whole; in roll each side migrates by its own arc length.
03After a ball rolls one lap along a circular path, does the net rotation angle depend on how fast it rolled? On the radius ratio?
Not on the speed (Ω cancels in the closed form — a geometric quantity, not a kinematic one); it depends on ρ/R: angle = 2π√(1+(ρ/R)²). Plus scale invariance: scaling the path and the ball together leaves the net orientation unchanged.
04Why does straight-line rolling have no holonomy while a circular path does?
Holonomy accumulates in «changes of the direction of motion»: in straight rolling the angular velocity's direction is constant and the net effect is an ordinary rotation about a horizontal axis; on a circular path the angular velocity's direction rotates with the path, becoming a constant screw motion in the rotating frame, leaving a net rotation about a tilted axis after one lap. Changes of direction are the fuel of holonomy.