Visual walkthrough — Control moment gyroscopes (CMG) — high torque, singularity

---

Step 1 — What is angular momentum, as a picture?

WHAT. Imagine a heavy disk (a flywheel) spinning fast. Physics gives it a single arrow that captures "how much spin, about which line, which way." We call this arrow the angular momentum and write it $\vec{h}$. The little arrow-hat over the $h$ just means "this is a vector — it has a direction, not only a size."

WHY an arrow and not just a number? Because spin has a direction: the disk spins about some axis. We point the arrow along that axis (right-hand rule: curl your right fingers with the spin, thumb points along $\vec{h}$). The length of the arrow is how much spin there is.

PICTURE. The disk lies flat; the burnt-orange arrow $\vec{h}$ sticks straight up through its centre.

---

Step 2 — Freeze the length, turn only the direction

WHAT. A CMG never speeds the wheel up or slows it down. It grabs the whole spinning wheel and tilts its axis. So the arrow $\vec{h}$ keeps its length $h$ but its head swings to a new direction.

WHY does keeping the length fixed matter? Because the entire magic of a CMG hinges on it. We are not going to grow the arrow (that would be a reaction wheel, which changes ${\omega }_w$). We only rotate an already-long arrow — that is cheap to do and, as we'll see, delivers a big effect.

PICTURE. The tip of $\vec{h}$ traces a circle of radius $h$. The arrow is a fixed-length hand of a clock; only its angle changes.

---

Step 3 — Newton's law for rotation: torque = change of $\vec{h}$

WHAT. The rotational version of "force changes momentum" is: torque changes angular momentum. In symbols: $\vec{\tau }=\frac{d\vec{h}}{dt}.$

WHY this tool, the derivative $\frac{d}{dt}$? We want to know the torque the CMG delivers. Torque is literally the rate at which the spin arrow changes. The symbol $\frac{d\vec{h}}{dt}$ answers exactly one question: "how fast, and in which direction, is the tip of $\vec{h}$ moving right now?" That is the precise question we need — no other tool answers "instantaneous change of a vector."

PICTURE. Two snapshots of the clock hand a heartbeat apart. The tiny green arrow from old tip to new tip is $d\vec{h}$; divide by the tiny time $dt$ and you get the torque arrow.

---

Step 4 — How does a rotating arrow's tip move? Enter $\vec{\Omega }\times \hat{s}$

WHAT. The gimbal spins the axis $\hat{s}$ about a fixed line called the gimbal axis $\hat{g}$, at a turning rate we call $\dot{\delta }$ ("delta-dot"). Here $\delta$ is the gimbal angle and $\dot{\delta }$ is how fast that angle grows. Package the turning as one arrow $\vec{\Omega }=\dot{\delta }\hat{g}$: it points along the axis you're turning about, and its length is the turning rate.

There is a universal fact about any unit arrow being rotated: $\frac{d\hat{s}}{dt}=\vec{\Omega }\times \hat{s}.$

WHY the cross product $\times$, and not multiply or add? We need the tip velocity of a spinning unit arrow. The tip moves on a circle: its speed is (turning rate)$\times$(radius of that circle), and its direction is perpendicular to both the turning axis $\hat{g}$ and the arrow $\hat{s}$. That "size = product of the two, direction = perpendicular to both" is precisely the job description of the cross product $\vec{\Omega }\times \hat{s}$. No plain multiplication gives a perpendicular direction; only the cross product does. That is why this tool and no other.

PICTURE. The gimbal axis $\hat{g}$ (teal, vertical). The spin arrow $\hat{s}$ leaning off it. Its tip rides a plum-coloured circle; the velocity arrow $d\hat{s}/dt$ is tangent to that circle — at right angles to both $\hat{g}$ and $\hat{s}$.

---

Step 5 — Assemble: the CMG torque law

WHAT. Put Step 3 and Step 4 together. Since $\vec{h}=h\hat{s}$ with $h$ frozen, differentiate: $\vec{\tau }=\frac{d\vec{h}}{dt}=h\frac{d\hat{s}}{dt}=h(\vec{\Omega }\times \hat{s})=\vec{\Omega }\times (h\hat{s})=\vec{\Omega }\times \vec{h}.$

WHY are we allowed to pull $h$ inside/outside? Because $h$ is a plain constant number (Step 2 froze it). A constant slides freely through a derivative and through a cross product. Every equality above is just "move a constant around."

PICTURE. The final triangle of arrows: $\vec{\Omega }$ (teal) up, $\vec{h}$ (orange) leaning, and their cross product $\vec{\tau }$ (plum) shooting out perpendicular to the plane they share.

---

Step 6 — Degenerate case: when the tilt does nothing ($\theta =0$)

WHAT. Look at $|\vec{\tau }|=h\dot{\delta }\sin \theta$ and push $\theta \to 0$: if you try to turn $\vec{h}$ about its own axis ($\hat{g}$ parallel to $\vec{h}$), then $\sin 0=0$ and the torque vanishes.

WHY? Spinning an arrow about the line it already points along moves its tip nowhere. No tip motion (Step 3) means no torque. This is the seed of trouble: a direction the CMG cannot push.

PICTURE. $\hat{g}$ laid flat along $\vec{h}$; the tip circle collapses to a point. Torque arrow gone.

---

Step 7 — Many CMGs: same picture, stacked into a Jacobian

WHAT. Each CMG $i$ contributes a torque column $\partial {\vec{h}}_i/\partial {\delta }_i={\hat{g}}_i\times {\vec{h}}_i$ — its own "which way can I push right now" arrow. Stack $n$ of them side by side into a matrix $A\in {\mathbb{R}}^{3\times n}$, the Jacobian. The total body torque is $\vec{\tau }=-A\dot{\vec{\delta }}$.

WHY a matrix? We have $n$ knobs (the gimbal rates ${\dot{\delta }}_i$) and want a $3$-D torque. A $3\times n$ matrix is exactly "combine $n$ inputs into a $3$-D output." To hit a commanded torque you invert it with the pseudo-inverse $A^{+}$.

PICTURE. Four little plane-arrows (one per CMG) fanned out; when they all flatten against a common line $\hat{u}$, the reachable set becomes flat — the singular wall of Step 6, now in a cluster.

---

The one-picture summary

Everything collapses to one chain: frozen-length arrow → turn it → tip moves perpendicular → that motion is the torque → when the tip can't move along a line, that line is a singularity.

Recall Feynman retelling — say it out loud in plain words

A CMG is a fast-spinning wheel whose "spin arrow" $\vec{h}$ has a fixed length — we never speed the wheel up or down. When a small motor tilts that arrow about a gimbal line $\hat{g}$, the tip of the arrow sweeps along a little circle. Newton's rotation law says torque is just how fast the tip moves, so the torque points where the tip is heading — which is sideways, at a right angle to both the tilt axis and the arrow. Because the arrow is already long, even a slow tilt whips its tip fast: big torque, tiny effort — that's the amplification. But if you ever tilt the arrow about its own direction, the tip goes nowhere and you get zero torque. Line up several such wheels and they can all lose the same direction at once: a wall called a singularity, which we tiptoe around with a damped (robust) inverse and torque-free null motion.

Recall Predict-then-verify checkpoints

Which tool gives the tip velocity of a rotating unit vector, and why that one? ::: The cross product $\vec{\Omega }\times \hat{s}$ — it alone returns a result perpendicular to both the turn axis and the arrow, with size = rate × radius. Why can we pull $h$ out of the derivative in Step 5? ::: Because $h$ is a frozen constant, and constants slide freely through derivatives and cross products. When does a single CMG produce zero torque? ::: When the gimbal axis is parallel to $\vec{h}$ ($\theta =0$), so $\sin \theta =0$ and the tip doesn't move. What geometric event makes a cluster singular? ::: All columns ${\hat{g}}_i\times {\vec{h}}_i$ become coplanar, sharing a common perpendicular $\hat{u}$ the cluster cannot torque along.