Spacecraft-Centred Frames and Rotations

RTN, VNC, LVLH and body axes; direction cosine matrices, Euler angles and quaternions

Mir, 25 June 1997

Commander Vasily Tsibliyev is flying an uncrewed Progress cargo ship into a docking with the Mir space station — by hand, by remote control, from inside Mir. It is a test of a manual backup system. All he has is a joystick and a black-and-white television picture from a camera on the Progress, looking back at the station. The picture shows him where Mir is in the camera's frame: up, down, left, right. It does not tell him how far away it is, or how fast he is closing. The radar that would have given range and rate has been switched off for the test. Mir's own modules block much of the view, and the station grows on the screen far later than anyone expects. The Progress, seven tonnes, hits the Spektr module at a few metres per second, punctures its hull and bends a solar array. The crew hear the hiss of escaping air, cut the cables and seal the hatch with seconds to spare. The frame he was given was the wrong one for the job. Nothing in this lesson is abstract.

Watch first · presented by Shoaib Aziz

The numbers in this film are the labs' own: a chaser 1 km above its target ends one orbit 9.424 km behind (3π ≈ 9.425); the same three turns in the reverse order land 26.1° apart; at 90° pitch two different angle sets give the same attitude; and mixing up VNC and RTN on an e = 0.6 orbit costs nothing at perigee or apogee but 6.2 m/s in between.

Planet-centred frames answer "where is it?". Operations ask different questions: which way is ahead? how far below the other vehicle am I? which thruster fires to slow down? For those you need frames that ride along with the spacecraft, rebuilt at every instant from its own position and velocity.

There are four in daily use. They are built from the same two vectors, r and v, and they differ only in which one they take as primary — which is exactly why people confuse them.

Frame	Axis 1	Axis 2	Axis 3	Used for
RTN (also RSW, RIC, Hill)	R̂ radial, out from the centre	T̂ transverse, in-plane, ⟂ R̂	N̂ orbit normal, along r × v	Relative motion, covariance, conjunction screening
VNC	V̂ along velocity	N̂ orbit normal	Ĉ = V̂ × N̂ (co-normal)	Specifying burns; drag and thrust directions
NTW	N̂ in-plane, ⟂ velocity	T̂ along velocity	Ŵ orbit normal	Vallado’s manoeuvre frame; same idea as VNC, relabelled
LVLH (CCSDS)	x̂ ≈ along-track	ŷ = −orbit normal	ẑ nadir, toward the Earth	Earth-pointing attitude (cameras, antennas, the ISS)

\hat{R} = \frac{\vec{r}}{r}, \quad \hat{N} = \frac{\vec{r}\times\vec{v}}{|\vec{r}\times\vec{v}|}, \quad \hat{T} = \hat{N}\times\hat{R}, \qquad \mathbf{M}_{\mathrm{RTN}} = \begin{bmatrix} \hat{R}^{\mathsf T} \\ \hat{T}^{\mathsf T} \\ \hat{N}^{\mathsf T} \end{bmatrix}, \quad \vec{x}_{\mathrm{RTN}} = \mathbf{M}_{\mathrm{RTN}}\,\vec{x}_{\mathrm{inertial}}

Stack the unit vectors as rows and you have the rotation matrix. Its transpose takes you back. Every local frame is built the same way.

On a circular orbit, T̂ and V̂ point in the same direction, so RTN and VNC are the same frame with the axes in a different order. On an eccentric orbit they are not. The velocity tilts away from the transverse direction by the flight-path angle γ — zero at perigee and apogee, largest in between:

\tan\gamma = \frac{e\sin\nu}{1 + e\cos\nu}, \qquad \hat{V} = \cos\gamma\,\hat{T} + \sin\gamma\,\hat{R}

For e = 0.6 at ν = 90°, γ = 31°. A burn “along-track” in one frame is 31° off in the other — lab 3 measures what that costs.

💡Covariance in the RTN frame

Orbit determination never gives you a point. It gives a position plus a covariance matrix P — the 3-D ellipsoid you are, say, 68% sure the satellite is inside. In inertial XYZ that matrix is full and unreadable. Rotated into RTN it nearly always becomes close to diagonal, and the physics jumps out: Radial error is small and stays small — altitude is well determined. Cross-track error is small and oscillates once per orbit. Along-track error is the big one, and it grows — roughly quadratically with time since the last tracking pass. A tiny error in period becomes a large error in when the satellite arrives. That is why every conjunction warning, every covariance in a CCSDS message, and every collision probability is stated in RTN.

\mathbf{P}_{\mathrm{RTN}} = \mathbf{M}\,\mathbf{P}_{\mathrm{XYZ}}\,\mathbf{M}^{\mathsf T}, \qquad \mathbf{P}_{\mathrm{XYZ}} = \mathbf{M}^{\mathsf T}\,\mathbf{P}_{\mathrm{RTN}}\,\mathbf{M}

A vector rotates with M once; a covariance rotates with M on both sides. The trace — total variance — is unchanged, because rotations do not create or destroy uncertainty.

🛰ISS debris-avoidance manoeuvres

The ISS performs a debris avoidance manoeuvre when the computed probability of collision with a tracked object exceeds about 1 in 10,000 — often with only hours of warning. That probability comes from the two objects' covariances, combined and projected onto the plane perpendicular to their relative velocity at closest approach. The along-track direction dominates the uncertainty, so when an alert arrives, the first operational question is rarely "how close?" but "how sure are we about when?". The answer lives in the T-component of an RTN covariance.

RTN is also where relative motion becomes intuitive — or rather, where it becomes counter-intuitive in a way you can learn. Put a chaser 1 km directly above a target, on its own circular orbit. Higher orbits are slower. After one revolution the chaser has fallen about

3\pi

km behind:

\Delta T_{\text{per orbit}} \approx -3\pi\,\Delta a

To catch up with something ahead of you, drop to a lower orbit. The rendezvous lesson in Unit 6 (Lesson 6-4) is built on this; Mir’s collision is what happens when your frame hides it.

💡The algebra of rotations

Every conversion in this unit is a direction cosine matrix (DCM): a 3×3 rotation C whose rows are one frame's axes written in the other. A rotation has strong properties, and checking them catches bugs: Orthonormal: CᵀC = I, so C⁻¹ = Cᵀ. Going backwards is free. Proper: det C = +1. A determinant of −1 means you built a mirror, not a rotation — usually a left-handed axis. Three degrees of freedom: nine numbers, six constraints. Order matters: composition is matrix multiplication, which does not commute. C₂C₁ means "do C₁ first". Attitude is the same mathematics, for the spacecraft's body frame. Engineers parameterise it in three common ways: Euler angles — three successive rotations about body axes. Twelve possible sequences: 3-1-3 (the orbital elements Ω, i, ω) and 3-2-1 (yaw, pitch, roll) are the famous ones. Intuitive, but every sequence has a singularity. Quaternions — four numbers with one constraint, no singularities, cheap to compose and to interpolate. The flight-software standard. q and −q describe the same attitude. The DCM itself — nine numbers, no singularity, wasteful but unambiguous.

\mathbf{C}_{321}(\psi,\theta,\phi) = \mathbf{R}_1(\phi)\,\mathbf{R}_2(\theta)\,\mathbf{R}_3(\psi), \qquad \theta = 90^\circ \;\Rightarrow\; \mathbf{C} \text{ depends only on } \psi - \phi

Gimbal lock: at 90° pitch, the first and last rotations act about the same axis, and one degree of freedom disappears. Apollo’s inertial platform had a “gimbal lock” warning light for exactly this.

Loading interactive…

⚠Hamilton and JPL quaternion conventions

There are two quaternion conventions in common use. The Hamilton convention (most robotics, ROS, Eigen) writes the scalar first and multiplies one way. The JPL convention (much of NASA's attitude literature) writes the scalar last and defines the product with the opposite handedness. Both are internally consistent. Mixed, they silently produce the inverse rotation. Add to that active versus passive rotations (rotating the vector versus rotating the axes), and the same four numbers can mean four different things. Every interface that passes an attitude must state: scalar first or last, which frame to which frame, and active or passive. The labs in this unit use passive (frame) rotations and scalar-last quaternions, and say so.

Here as he walked by on the 16th of October 1843 Sir William Rowan Hamilton in a flash of genius discovered the fundamental formula for quaternion multiplication $i^2 = j^2 = k^2 = ijk = -1$ & cut it on a stone of this bridge.

– Plaque on Broom Bridge, Dublin

💡Star trackers and the link back to the ICRS

How does a spacecraft know its own attitude? With a star tracker: a camera that photographs the sky, matches the star pattern against an onboard catalogue, and outputs a quaternion from the inertial frame to its own body axes. The catalogue is Hipparcos or Gaia — which means the stars' coordinates are in the ICRS, tied to the quasars of lesson 1. So the attitude of every modern spacecraft is ultimately measured against galaxies billions of light-years away, through a chain of frames this unit has now built end to end.

↓But there is a problem

You can now move between celestial, terrestrial and spacecraft frames at will — around Earth. Leave Earth and the questions change. Where is the Moon's prime meridian, and why are there two of them half a mile apart? Which way is north on Uranus, which is lying on its side and spinning backwards? What frame do you use for a spacecraft parked near a point in space that is not attached to any planet at all? Mission designers answer these every day, and their answers are a little stranger than you would expect.