Attitude estimation — triad method (two vector measurements): Question bank

3.5.12 · D5 · Physics › Guidance, Navigation & Control (GNC) › Attitude estimation — triad method (two vector measurements)

Ye conceptual traps hain, calculators nahi. Har line ek question ::: answer reveal hai — peeking se pehle apna jawab out loud bolne ki koshish karo. Har answer mein reasoning milegi, sirf haan/na nahi.

Agar koi symbol unfamiliar lage, toh parent note se rebuild karo: the TRIAD topic note. Ye prerequisites bhi kholne wale hain: Cross Product and Right-Handed Frames, Rotation Matrices and SO(3).

Neeche har trap $\hat{t}_{1}, \hat{t}_{2}, \hat{t}_{3}$ triad axes aur matrices $M_{r}, M_{b}$ ke baare mein baat karta hai. Yahan exactly likha hai ki har symbol ka matlab kya hai, taaki questions mein koi undefined cheez chhup ke na aa jaaye. Har line padhte waqt picture dekho.

Figure — Attitude estimation — triad method (two vector measurements)

Definition Teen triad axes (dono frames mein same tarike se build hote hain)

Shuru karo do unit measured directions se ( $\hat{r}_{1}, \hat{r}_{2}$ reference frame mein, ya $\hat{b}_{1}, \hat{b}_{2}$ body frame mein). $\hat{a}_{1}, \hat{a}_{2}$ ko "koi bhi pair" ke liye likhein, toh teen orthogonal unit axes hain:

\hat{\mathbf t}_2=\frac{\hat{\mathbf a}_1\times\hat{\mathbf a}_2}{\lVert\hat{\mathbf a}_1\times\hat{\mathbf a}_2\rVert},\qquad \hat{\mathbf t}_3=\hat{\mathbf t}_1\times\hat{\mathbf t}_2$$ - $\hat{\mathbf t}_1$ — **pehli, poori tarah trusted** direction, seedha copy ki gayi. - $\hat{\mathbf t}_2$ — ek **unit** vector jo dono inputs ke perpendicular hai (cross product ka yehi toh poora point hai), *normalized* taaki iska length exactly $1$ ho. - $\hat{\mathbf t}_3$ — **bacha hua** axis, right-handed set banana ke liye force kiya gaya.

Recall

$det A = + 1$ kyun hai (taaki $A$ ek real rotation ho, reflection nahi) Do facts kaam karte hain. (1) Determinant-multiplication rule: $det (X Y) = det X det Y$ , aur $det (M^{T}) = det M$ . (2) Har triad matrix ka $det = + 1$ hota hai: iske columns orthonormal hain isliye $det = \pm 1$ , aur $\hat{t}_{3} = \hat{t}_{1} \times \hat{t}_{2}$ build karna (usi order mein) right-handed sign $+ 1$ force karta hai. Tab $det A = det (M_{b}) det (M_{r}^{T}) = det (M_{b}) det (M_{r}) = (+ 1) (+ 1) = + 1.$ Isi tarah $A A^{T} = M_{b} M_{r}^{T} M_{r} M_{b}^{T} = M_{b} M_{b}^{T} = I$ , isliye $A$ ek proper rotation hai ✅.

True ya false — justify karo

Do unit vectors ek full 3D attitude determine karne ke liye kaafi hain.

True, agar woh non-parallel hain. Ek direction ek spin degree of freedom free chodh deta hai; ek doosra independent direction us aakhri angle ko lock kar deta hai, jisse teeno rotational degrees of freedom mil jaate hain.

TRIAD dono measurements ko symmetrically use karta hai.

False.

\hat{b}_{1}

ek exact axis banta hai, lekin sirf

\hat{b}_{1} \times \hat{b}_{2}

ki direction bachti hai —

\hat{b}_{2}

ka precise angle discard ho jaata hai. Slot 1 aur 2 ko swap karna generally alag

A

deta hai.

Matrix

A = M_{b} M_{r}^{T}

guaranteed hai ki ek proper rotation hogi.

True, agar tumne har triad axis ko normalize kiya ho. Tab

M_{b}, M_{r}

orthonormal hain

det = + 1

ke saath, isliye

A A^{T} = I

aur

det A = + 1

— Rotation Matrices and SO(3) mein rotation ki defining properties.

Agar sensor 2 sensor 1 se zyada noisy hai, tab bhi sensor 1 ko slot 1 mein rakhna chahiye.

True. Slot 1 exactly honor kiya jaata hai, isliye apna sabse accurate sensor wahan rakho; slot 2 ka angular noise cross-product construction se kaafi had tak throw away ho jaata hai.

TRIAD woh

A

dhundtha hai jo least-squares sense mein dono vectors ko best fit kare.

False. Ye Wahba's Problem / QUEST algorithm hai. TRIAD koi optimization nahi karta — woh ek vector par poora trust karta hai aur doosre ko sirf plane ki orientation ke liye use karta hai.

\hat{b}_{1}

aur

\hat{b}_{2}

ke beech ka angle

\hat{r}_{1}

aur

\hat{r}_{2}

ke beech ke angle ke barabar hona chahiye taaki TRIAD koi answer de sake.

False. Noise un angles ko alag bana deta hai, phir bhi TRIAD ek valid rotation return karta hai — precisely kyunki woh

\hat{b}_{2}

ka angle ignore karta hai aur sirf uski direction rakhta hai.

\hat{t}_{3} = \hat{t}_{1} \times \hat{t}_{2}

\hat{t}_{2} \times \hat{t}_{1}

se replace kiya ja sakta hai bina kisi consequence ke.

False. Cross product ko reverse karna

\hat{t}_{3}

ka sign flip kar deta hai, frame ko left-handed bana deta hai (

det = - 1

) — ek improper rotation (ek reflection), jo ek physical attitude nahi hai.

\hat{t}_{2}

ko normalize karna optional hai agar inputs already unit vectors hain.

False. Unit inputs ke liye bhi,

∥ \hat{r}_{1} \times \hat{r}_{2} ∥ = sin θ \neq = 1

jab tak

θ = 9 0^{\circ}

na ho. Normalization skip karna orthonormality tod deta hai aur

M^{- 1} \neq = M^{T}

ho jaata hai.

TRIAD theek kaam karta hai jab do reference vectors Sun aur ek second baad measure kiya geomagnetic field hoon.

True in principle, jab tak woh angle mein well separated hoon. Sun aur field typically non-parallel hote hain, yehi wajah hai ki Sun Sensors and Magnetometers ek classic TRIAD pair hain.

Error dhundho

"Main

A = M_{b} M_{r}^{- 1}

directly

M_{r}

ko numerically invert karke compute karunga — same cheez hai."

Result same hai sirf tabhi jab

M_{r}

orthonormal ho, us case mein

M_{r}^{- 1} = M_{r}^{T}

exactly hota hai aur transpose sasta aur exactly stable hai. Ek matrix ka numerical inversion jiske baare mein tum jaante ho ki woh orthonormal hai, effort waste karta hai aur error introduce kar sakta hai.

"Mera cross product

(0, 0, 0.02)

aaya, main isse

(0, 0, 1)

par normalize karke continue karunga."

Near-zero cross product matlab hai ki do vectors nearly parallel hain (

sin θ \approx 0.02

). Normalize karna tiny noisy tail ko ek full unit vector mein amplify kar deta hai, isliye doosra axis — aur uske baare mein poora attitude — noise se dominate ho jaata hai.

"Maine reference triad

\hat{r}_{1}, \hat{r}_{2}

se build kiya lekin body triad

\hat{b}_{2}, \hat{b}_{1}

se (swapped order)."

Recipe dono sides par identical honi chahiye. Sirf ek side par dono vectors ke roles swap karna matlab hai ki

M_{r}

aur

M_{b}

alag physical triads describe karte hain, isliye

A = M_{b} M_{r}^{T}

meaningless hai.

"Maine

A M_{b} = M_{r}

set kiya aur

A

solve kiya."

Ulta hai. Attitude

A

reference → body map karta hai:

A \hat{r}_{i} = \hat{b}_{i}

, isliye

A M_{r} = M_{b}

. Tumhara equation solve karna

A^{T}

deta hai — inverse rotation.

"Vectors ko unit vectors hone ki zaroorat nahi; TRIAD sirf direction ki parwah karta hai."

Construction zaroori hai ki unit vectors hoon:

\hat{t}_{1} = \hat{b}_{1}

directly ek orthonormal matrix ke column mein copy hota hai. Wahan ek non-unit vector

M^{- 1} = M^{T}

tod deta hai. Pehle raw measurements normalize karo.

"Mujhe

det A = - 1

mila; rotation ke kaafi kareeb hai."

det A = - 1

ek reflection hai, rotation nahi — ye handedness flip karta hai. Almost hamesha matlab hai ki koi cross-product sign reverse ho gayi ya koi frame left-handed build hua. Ye ek hard error hai, roundoff nahi.

Why questions

Cross product kyun hai, average ya difference kyun nahi,

\hat{t}_{2}

build karne ke liye?

Cross product woh ek operation hai jiska output dono inputs ke perpendicular hota hai,

\hat{t}_{1}

se orthogonality guarantee karta hai. Dekho Cross Product and Right-Handed Frames. Average ya difference plane mein rehte hain aur koi nayi orthogonal axis nahi dete.

TRIAD doosre measurement ka angle kyun "throw away" karta hai?

Kyunki

\hat{t}_{2}

sirf

\hat{b}_{1} \times \hat{b}_{2}

ki direction par depend karta hai, uske magnitude par nahi — isliye

\hat{b}_{2}

mein ek chhoti si angular error sirf plane ko thoda tilt karti hai, aur exact angle kabhi

A

mein enter nahi karta. Ye deliberate noise-rejection choice hai.

Hume specifically

det A = + 1

kyun chahiye, sirf

A A^{T} = I

kyun nahi?

A A^{T} = I

det = \pm 1

allow karta hai;

- 1

case ek reflection hai jo koi rigid body physically perform nahi kar sakta.

det = + 1

require karna

A

ko genuine rotations tak restrict karta hai, group Rotation Matrices and SO(3).

Ek single vector measurement attitude ke liye insufficient kyun hai?

Ek single direction 3 rotational degrees of freedom mein se 2 pin down karta hai — body us direction ke around freely spin karta reh sakta hai aur koi single measurement change nahi hoga. Doosra vector us last spin ambiguity ko remove karta hai.

9 0^{\circ}

ke paas separation dono vectors ke liye best kyun hai?

Cross-product magnitude

sin θ

hai, jo

θ = 9 0^{\circ}

par peak karta hai. Ek bada, well-conditioned cross product matlab hai ki plane ki orientation noise ke liye robust hai;

0^{\circ}

18 0^{\circ}

ke paas ye zero ki taraf collapse ho jaata hai aur noise dominate kar leti hai.

TRIAD QUEST algorithm / Davenport q-method se kyun haarta hai jab dono sensors comparably accurate hoon?

TRIAD ek sensor ka angle poori tarah sacrifice kar deta hai. Jab dono sensors useful information carry karte hain, ek optimal least-squares fit (Wahba's Problem) sab kuch use karta hai aur kam-error estimate deta hai; TRIAD achha data throw away kar raha hai.

Attitude yahan quaternion ki jagah matrix ke roop mein kyun express karein?

Triad construction naturally column-by-column matrix form mein hai. Result ko baad mein Quaternion Attitude Kinematics mein propagation ke liye ya Kalman Filter — Attitude mein filtering ke liye quaternion mein convert kiya ja sakta hai.

Edge cases

Kya hota hai jab

\hat{r}_{1}

aur

\hat{r}_{2}

exactly parallel hoon?

Cross product zero vector hai,

sin θ = 0

, aur

\hat{t}_{2}

undefined hai (zero se division). Shared axis ke baare mein attitude fundamentally undetermined hai — koi bhi method, sirf TRIAD nahi, ise recover nahi kar sakta.

Kya hota hai jab do vectors parallel ke paas aate hain lekin bilkul nahi?

TRIAD phir bhi ek answer return karta hai, lekin tiny denominator

sin θ

\hat{t}_{2}

mein measurement noise amplify karta hai (aur isliye pure spin angle mein bhi) — estimate continuously degrade hoti hai aur true collinearity se kaafi pehle untrustworthy ho jaati hai.

Kya agar

\hat{b}_{1}

perfect hai lekin

\hat{b}_{2}

complete garbage hai (random direction)?

Axis

\hat{t}_{1} = \hat{b}_{1}

ab bhi exact hai, isliye us axis ke baare mein attitude anchored rehta hai; lekin us axis ke baare mein spin angle, jo

\hat{b}_{2}

ke plane se set hota hai, garbage hai. Tumhare paas do achhe DOF aur ek ruined milta hai.

Kya agar tumhare paas do nahi, teen vector measurements hoon?

Plain TRIAD sirf do use karta hai — best pair chuno, ya better, QUEST algorithm / Davenport q-method par switch karo, jo teeno ko optimally fuse karta hai data discard karne ki jagah.

Jab measurements exactly noise-free hoon tab estimate error kya hai?

Zero. Noiseless case mein reference aur body triads exact rotations of each other hain, isliye

A = M_{b} M_{r}^{T}

true attitude perfectly reproduce karta hai,

A \hat{r}_{i} = \hat{b}_{i}

dono

i

ke liye satisfy karta hai.

Kya agar dono sensors ki accuracy identical ho — kya slot assignment matter karta hai?

Numerically dono orderings alag

A

's dete hain, lekin na hi "zyada correct" hai. Kyunki koi bhi sensor full trust ka deserving nahi, ye exactly woh regime hai jahan tum TRIAD ko ek optimal method ke liye abandon karo jo unhe fairly treat kare.

Recall Trap themes ka one-line summary

Q1 Asymmetry (slot 1 exact, slot 2 direction-only). ::: TRIAD ek vector par fully trust karta hai. Q2 Fragility. ::: $sin θ$ se divide karna parallel ke paas blowup karta hai. Q3 Validity. ::: Sab kuch normalize karo; $det A = + 1$ demand karo, warna ye reflection hai.