Control Moment Gyroscopes: How Spacecraft Turn Without Fuel

The core idea: exchange momentum, don't expel it

In free-fall there is nothing to push against, so a spacecraft cannot simply "steer." Its total angular momentum H is conserved unless an external torque acts (gravity gradient, aerodynamic drag, solar pressure, magnetic torque). A CMG exploits this: the vehicle + wheel form one closed system. If the wheel's angular-momentum vector is rotated one way, the vehicle's body must rotate the opposite way so the total stays fixed. This is a momentum-exchange actuator — momentum is traded internally, not thrown overboard as with a thruster.

Each CMG carries a rotor spinning at a constant high speed, giving it a fixed-magnitude angular momentum H = I⋅ω pointed along the spin axis. A gimbal motor slowly rotates that whole spinning assembly, so the direction of H sweeps through space. The rate of change of that vector is a torque, and by Newton's rotational law that torque is delivered straight into the spacecraft structure.

The physics: τ = ω_gimbal × H, and why the torque is amplified

The reaction torque a single-gimbal CMG applies to the vehicle is the cross product of the gimbal rate and the wheel momentum:

τ = ω_gimbal × H_wheel

Its magnitude, when the two vectors are perpendicular, is simply τ = ω_gimbal ⋅ H_wheel, and it points perpendicular to both the spin axis and the gimbal axis — a defining, non-intuitive feature of gyroscopics. Tilt the spinning wheel about one axis and the craft turns about a third, orthogonal axis.

Here is why CMGs beat spinning-up a wheel. Take the ISS-class rotor: H ≈ 4,881 N⋅m⋅s (220 kg wheel at 6,600 rpm). Command a modest gimbal rate of just ω_gimbal ≈ 3°/s ≈ 0.052 rad/s. The output torque is τ = 0.052 × 4,881 ≈ 255 N⋅m. To make that same torque with a reaction wheel you would have to angularly accelerate the same 220 kg rotor — enormous motor power and it saturates in seconds. The CMG instead redirects a momentum reservoir that is already spinning, so the gimbal motor does almost no work against the spin; it only steers the vector. That is the torque amplification, commonly 10× to 100×.

Single-gimbal, double-gimbal, and how you build an array

• Single-gimbal CMG (SGCMG): one gimbal axis per unit. Highest torque amplification and simplest hardware, but each unit produces torque in only one plane. You need ≥3 (usually 4 for redundancy) arranged so their torque directions span 3-D space.
• Double-gimbal CMG (DGCMG): the wheel is nested in two gimbals, so one unit can point H anywhere on a sphere and produce torque in two axes. More versatile (Skylab and the ISS use these), but heavier, slower, and lower peak torque.
• Arrays: Multiple SGCMGs are grouped in a pyramid (typically 4 units on the faces of a pyramid with a ~54.7° skew angle) or a roof/scissored-pair geometry. A steering law computes the gimbal rates δ̇ that produce the demanded body torque, usually by inverting the array Jacobian: τ = A(δ)⋅δ̇, solved as δ̇ = A⁺⋅τ with a pseudo-inverse.

The ISS carries four DGCMGs on the Z1 truss; three are enough to control attitude, the fourth is redundancy. Together they store roughly ~20,000 N⋅m⋅s of momentum. Skylab flew the first operational set — three DGCMGs — in 1973.

The catch: singularities and momentum saturation

The signature failure mode of a CMG array is a geometric singularity. As the gimbals rotate to different angles, there are configurations where all the individual torque vectors become coplanar or collinear — the array can produce no torque along one particular direction, no matter how the gimbals move. Mathematically the Jacobian A(δ) loses rank and its pseudo-inverse blows up, commanding impossibly large gimbal rates. Steering laws must actively avoid or escape these states, using methods such as Singularity-Robust (SR) inverse steering (which adds a small damping term and accepts a tiny torque error) or null-motion that reshuffles gimbals without net torque.

The second limit is momentum saturation, shared with reaction wheels. External disturbances (on the ISS, mainly gravity-gradient and aerodynamic torque) keep pushing momentum into the array. Once the wheels' combined momentum is maxed, the CMGs can absorb no more. The stored momentum must be dumped — "desaturated" or "unloaded" — using an external torque source: RCS thrusters, magnetic torque rods (interacting with Earth's field), or, cleverly on the ISS, periodic gravity-gradient maneuvers (the Zero-Propellant Maneuver, first flown 2006) that steer the station into an attitude where gravity itself bleeds momentum back out for free.

Real hardware, real numbers

ISS DGCMG (built by L3/Honeywell): a 220 kg (≈490 lb) steel rotor, ~1 m assembly, spun at 6,600 rpm by a brushless DC motor, storing ~4,881 N⋅m⋅s and delivering up to ~258 N⋅m. The units run for years; CMG-1 failed in 2002 (a spin-bearing/vibration fault) and was replaced by spacewalk in 2005 — a stark reminder that the spin bearing, running continuously at thousands of rpm in vacuum, is the life-limiting component.

Small/agile satellites: Airbus/SSTL and Honeywell make compact SGCMGs in the 4–15 N⋅m⋅s class for Earth-imaging platforms (e.g. Pléiades, WorldView-class agile satellites), enabling slew rates well above 1°/s so a single pass can shoot multiple targets. Honeywell's M50/M95 lines and Airbus "CMG 15-45" families are typical catalog units.

Skylab (1973): three double-gimbal CMGs, each with a ~65 kg rotor near 9,000 rpm, holding roughly 3,100 N⋅m⋅s — the first spacecraft to rely on CMGs for primary attitude control, which stretched its limited thruster propellant across the whole mission.

Where CMGs are used — and where they are not

CMGs earn their place wherever a vehicle is large, needs agility, or must run for years without spending fuel on pointing:

• Space stations: ISS (four DGCMGs), Skylab, and Mir all used CMGs to hold attitude cheaply for their entire lifetimes.
• Space telescopes: Hubble uses four reaction wheels, not CMGs — an important nuance (see below) — but large agile observatories and defense sats use CMGs when fast retargeting matters.
• Agile Earth-imaging satellites: CMG arrays let a satellite roll 30–45° in seconds to hit off-nadir targets, the key competitive spec for commercial imagery.

They are not used on most CubeSats or coarse-pointing smallsats: the control law is complex, singularity handling costs software effort, and a simple reaction wheel or magnetorquer is cheaper, lighter, and good enough. CMGs win only when you need much more torque per kilogram than a reaction wheel can give.

A common misconception and a subtle pitfall

Misconception: "The gyroscope's spin resists motion and that is what holds the craft." Not quite. A CMG does not work by rigidity of a spinning wheel; it works by rotating the momentum vector to make a controllable torque. A reaction wheel (which changes spin speed) and a CMG (which changes spin direction) are both momentum-exchange devices — the difference is purely which property of H = I⋅ω you modulate: its magnitude (reaction wheel) or its direction (CMG). Hubble uses reaction wheels; the ISS uses CMGs.

Subtle pitfall: a CMG's torque direction depends on the current gimbal angle, so the effective control authority is not constant — it shrinks and even vanishes near singularities. A naïve linear controller that assumes fixed torque authority can command the array straight into a singularity and stall the slew mid-maneuver. Robust steering (SR-inverse, null-motion redistribution, or global steering laws) is not optional on a real vehicle — it is what separates a textbook CMG from one that actually flies.

Control Moment Gyroscope vs Reaction Wheel — two momentum-exchange actuators compared

Attribute	Control Moment Gyroscope	Reaction Wheel
How torque is made	Tilt a constant-speed wheel's spin axis (gimbal)	Change a wheel's spin speed (accelerate/decelerate)
Torque per watt	High — torque amplification 10–100×	Low — torque = wheel inertia × angular acceleration
Peak torque	Tens to hundreds of N⋅m (e.g. ~258 N⋅m ISS)	Typically 0.01–1 N⋅m (small sats), a few N⋅m large
Control complexity	High — nonlinear, singularity avoidance needed	Low — near-linear, simple to steer
Failure/limit mode	Geometric singularities (torque direction lost)	Wheel saturation at max rpm
Best fit	Large/agile craft: ISS, Hubble, imaging sats	Small sats, CubeSats, coarse pointing