Euler angles

The formula

The aerospace standard is the intrinsic z-y′-x″ sequence: yaw ψ about the vertical, then pitch θ about the new lateral axis, then roll φ about the resulting longitudinal axis. Stacking the three elemental rotations gives the direction cosine matrix that takes a vector from body axes to the local navigation frame (NED):

R = Rz(ψ)·Ry(θ)·Rx(φ)

    [ cψ·cθ    cψ·sθ·sφ − sψ·cφ    cψ·sθ·cφ + sψ·sφ ]
R = [ sψ·cθ    sψ·sθ·sφ + cψ·cφ    sψ·sθ·cφ − cψ·sφ ]
    [ −sθ      cθ·sφ               cθ·cφ            ]

Reading: three angles, applied in a fixed order, compress into one 3×3 rotation matrix — nine numbers, but only three degrees of freedom, with det R = 1 and R·Rᵀ = I always.

Going the other way, angles from a known matrix:

θ = −asin(R₃₁)
φ = atan2(R₃₂, R₃₃)
ψ = atan2(R₂₁, R₁₁)        valid only while cos θ ≠ 0

Reading: pitch comes straight off one element; roll and yaw need atan2 to land in the right quadrant. The cos θ ≠ 0 caveat is not fine print — it is the whole gimbal-lock story.

The kinematic equations, which turn measured body rates (p, q, r) from the gyros into Euler-angle rates:

φ̇ = p + (q·sin φ + r·cos φ)·tan θ
θ̇ = q·cos φ − r·sin φ
ψ̇ = (q·sin φ + r·cos φ)/cos θ

Reading: body rates are not the same thing as angle rates. The tan θ and 1/cos θ terms blow up as pitch approaches ±90° — at θ = 89°, a 1 rad/s body yaw rate demands a 57.3 rad/s ψ̇. The math is telling you the parameterization is failing, not the vehicle.

Where you meet it

The flight-test telemetry room. Attitude comes down the stream as roll, pitch, heading because that is what a human can sanity-check at a glance. The flight computer almost certainly ran quaternions internally; Euler angles are the display format, converted at the last step for the strip charts and the pilot's ADI.
The seeker or camera gimbal test stand. A two- or three-axis gimbal is a physical Euler sequence — each ring is one elemental rotation. When the outer and inner axes line up, the mechanism itself loses a degree of freedom, and the rate loop commands whip toward infinity trying to track through the pole. That is gimbal lock as hardware, not as math.
The GNC design review. Someone presents an attitude interface and the first question from the back of the room is always the same: which sequence, intrinsic or extrinsic, degrees or radians, body-to-nav or nav-to-body. Five questions, thirty-two wrong combinations, and every ICD that skips them costs an integration week later.
The vibration or motion-platform lab. Six-DOF shaker tables and motion bases take pose commands as three translations plus roll-pitch-yaw. Feed one a set of angles built with the wrong order convention and the fixture goes somewhere legal, plausible, and wrong.

How it works

An orientation in 3D has exactly three degrees of freedom, and Euler's insight was that any orientation can be reached by three rotations about coordinate axes taken in sequence. There are twelve usable sequences: six that repeat the first axis (z-x-z, x-y-x, and so on — the "proper" Euler angles, native to spinning tops and orbital elements) and six that use all three axes (z-y-x, x-y-z, and friends — the Tait-Bryan angles, native to vehicles) ^[1]. None of them is "the" Euler angles. A tuple of three numbers means nothing until the sequence is stated.

Order matters because rotations don't commute. Yaw 90° then pitch 90° puts the nose somewhere entirely different than pitch 90° then yaw 90°. It isn't only a large-angle effect: swap the order of three 5° rotations and the resulting attitudes differ by about 0.75° — far beyond the pointing budget of any antenna, seeker, or star tracker you'll ever test.

The singularity is the defining flaw. At θ = ±90° the first and third rotation axes become collinear, and only the combination of yaw and roll is observable — at θ = +90°, the sets (ψ = 40°, φ = 10°) and (ψ = 70°, φ = 40°) produce the identical rotation matrix, because only ψ − φ survives. One degree of freedom is gone. Nothing is wrong with the physical orientation; straight up is a perfectly good attitude. The map from angles to orientations has a pole there, the same way longitude is meaningless at the North Pole. Any attitude estimator, autopilot, or animation tool built on Euler angles will do something ugly passing near the pole: angles that slew hundreds of degrees between samples, rates that spike, interpolation that takes the long way around. The standard fix is to run quaternions or the full DCM internally — both are singularity-free — and convert to Euler angles only for display ^[1]^[5].

The mistakes that actually bite, in rough order of frequency. One: assuming the other party's convention. Aerospace ZYX intrinsic, robotics XYZ, orbital mechanics 3-1-3 — all answer to the name "Euler angles." Two: the transpose. Body-to-nav and nav-to-body matrices differ by a transpose, and a transposed DCM still passes every orthonormality check while pointing your vector the wrong way. Three: using atan(y/x) instead of atan2(y, x) in the extraction and losing two quadrants. Four: averaging or interpolating angles across the ±180° wrap, which turns a 2° heading change into a 358° one. Five: treating (φ̇, θ̇, ψ̇) as if it were (p, q, r) — legitimate only for small angles and small rates, which is exactly the linearization Bryan used for aircraft stability derivatives, and exactly why it fails in aerobatic or tumbling flight.

Small-angle form, for what it's worth: at first order the sequence stops mattering and R ≈ I + [small skew matrix], which is why attitude-error states in Kalman filters are almost always three small angles even when the whole attitude is a quaternion.

History

Leonhard Euler (1707–1783) spent the 1760s taking rigid-body motion apart. His 1765 Theoria motus corporum solidorum split the general motion of a solid into a translation plus a rotation, and his study of the angles now carrying his name grew out of the precession of the equinoxes — an astronomy problem, not an aviation one ^[2]. A 1767 Berlin Academy memoir worked out the motion of a body turning about a moving axis ^[1]. He did much of this while going blind; after the failed cataract surgery of 1771 left him with no sight at all, he went on to produce roughly half of his life's output by dictation ^[2].

The vehicle-friendly variant came from two very different men. Peter Guthrie Tait (1831–1901), the Scottish physicist who co-wrote the Treatise on Natural Philosophy with Kelvin and spent decades championing quaternions as the future of mathematical physics, handled rotation sequences in that framework ^[1]^[3]. George H. Bryan (1864–1928) then carried the three-distinct-axes convention into the new problem of aircraft stability, publishing Stability in Aviation in 1911 — the book that set up the airplane equations of motion in the form flight dynamics texts still use ^[1]^[4]. The pairing has a dry irony: the angles named partly for Tait are the ones his beloved quaternions exist to replace.

The singularity earned its reputation in July 1969. The Apollo inertial platform used three gimbals; the designers knew a fourth would remove the lock condition and rejected it for size and simplicity, settling for a warning light at 70° and an IMU freeze at 85° that forced a manual star realignment ^[5]^[6]. Michael Collins, orbiting alone in the command module after Eagle had landed on the Sea of Tranquility, wrestling the platform around the restricted zone as he tried to spot the LM on the surface, radioed Houston: "How about sending me a fourth gimbal for Christmas?" ^[5]^[6]. He didn't get one. The industry got quaternions instead.

Related tools

/tools/convert-angle
/tools/moment-of-inertia-shapes
/tools/antenna-gain-beamwidth
/tools/orbital-velocity-period

Sources

Written by HE in our own words from the cited sources — engineering judgment included, your stamp still required. All entries →