Converting between DCM, quaternions, Euler angles: Question bank

3.5.10 · D5 · Physics › Guidance, Navigation & Control (GNC) › Converting between DCM, quaternions, Euler angles

Ye concept traps hain, calculators nahi. Har line ek claim hai, ek error hai, ek "kyun" hai, ya ek edge case hai. Answer cover karo, zor se bol ke commit karo, phir reveal karo. Agar tumhara reason sirf "yes/no" hai, to tumne jawab nahi diya — justification hi asli cheez hai.

Parent: 3.5.10 · Converting between DCM, quaternions, Euler angles. Ye prerequisites dobara kholne laayak hain: Euler's Rotation Theorem, Direction Cosine Matrix Properties, Gimbal Lock, Rotation Sequences (3-2-1 vs 3-1-3).

Notation jo tumhe traps se pehle chahiye

Definition Teen "languages" aur unke symbols

Neeche jo kuch bhi hai wo ek hi fact describe karta hai: ek body frame kaise rotate hota hai ek reference frame ke relative. Isko likhne ke teen tarike hain:

DCM = Direction Cosine Matrix, likha jaata hai $C$ : $3 \times 3$ numbers ka ek grid. $C_{ij}$ matlab row $i$ , column $j$ wali entry. Isko teen numbers ke ek column (ek vector $v$ , space mein ek arrow) pe apply karne se wo arrow rotate hota hai: $v^{'} = C v$ . $I$ identity matrix hai (diagonal pe $1$ s, baaki $0$ s) — "kuch mat karo" wala rotation.
Quaternion $q$ : chaar numbers jo scalar-first likhe jaate hain $q = (q_{0}, q)$ jahan $q_{0}$ scalar part hai aur $q = (q_{1}, q_{2}, q_{3})$ vector part hai. To $q = (q_{0}, q_{1}, q_{2}, q_{3})$ . Ye ek vector ko sandwich product $v \mapsto q v q^{- 1}$ se rotate karta hai (baayein $q$ se multiply karo aur daayein uske inverse $q^{- 1}$ se).
Euler angles $(ϕ, θ, ψ)$ : teen human-readable angles — roll $ϕ$ , pitch $θ$ , yaw $ψ$ — jo aerospace 3-2-1 (Tait–Bryan) sequence mein apply hote hain: $z$ ke baare mein yaw $ψ$ , phir naye $y$ ke baare mein pitch $θ$ , phir naye $x$ ke baare mein roll $ϕ$ . (Alag order, jaise 3-1-3, ek alag convention hai — dekho Rotation Sequences (3-2-1 vs 3-1-3).)

Ek aur tool: $[\overset{e}{^}]_{\times}$ axis $\overset{e}{^}$ ki cross-product matrix hai — wo $3 \times 3$ matrix jiske liye $[\overset{e}{^}]_{\times} v = \overset{e}{^} \times v$ . Ye antisymmetric hai ( $[\overset{e}{^}]_{\times}^{⊤} = - [\overset{e}{^}]_{\times}$ ).

Is page ke har trap mein body-from-inertial convention aur 3-2-1 sequence assume ki gayi hai jab tak line kuch aur na kahe. Neeche wali picture teen languages ko ek geometric rotation se pin karti hai taaki traps ke paas point karne ke liye kuch ho.

Figure — Converting between DCM, quaternions, Euler angles

True ya false — justify karo

Ek rotation matrix aur uska transpose ek hi physical orientation describe karte hain.

False.

C

aur

C^{⊤}

inverse rotations hain (body-from-inertial vs inertial-from-body). Ye sirf identity ya

180°

turn ke liye ek jaise hote hain; warna galat wala use karne se har mapping reverse ho jaati hai. Dekho Direction Cosine Matrix Properties.

Quaternions

q

aur

- q

opposite axes pe point karte hain, isliye ye opposite rotations hain.

False. Rotation map

v \mapsto q v q^{- 1}

q

mein quadratic hai, isliye dono minus signs cancel ho jaate hain aur dono identical DCM dete hain — same axis, same angle.

q

ko negate karna

Φ \to Φ + 360°

hai, jo koi change nahi hai. (Double-cover figure dekho.)

Nine real numbers ki ek

3 \times 3

matrix koi bhi rotation freely encode kar sakti hai.

False. Ek DCM orthonormal hona chahiye determinant

+ 1

ke saath: ye

6

constraints hain, sirf

3

free numbers bachte hain — rotation ki asli dimension. Nine "free" numbers shears aur reflections allow kar dete.

Euler angles mein singularity isliye hai kyunki physical rotation khud undefined ho jaata hai.

False. Rotation bilkul theek hai; sirf coordinate chart fail hoti hai. Pitch

\pm 90°

pe do gimbal axes align ho jaate hain to yaw aur roll separate nahi ho sakte — ye representation ki kami hai, motion ki nahi. Dekho Gimbal Lock aur neeche wala gimbal figure.

Quaternion set mein kabhi singularity nahi hoti, isliye ye human-readable attitude perfectly report kar sakta hai.

False (half-true). Quaternions store aur propagate karne ke liye singularity-free hain, lekin ye human-readable nahi hain — roll/pitch/yaw report karne ke liye tumhe phir bhi Euler mein convert karna padta hai, jo singular hota hai.

Kyunki

q_{0} = cos (Φ/2)

hai,

0°

ka rotation aur

360°

ka rotation same quaternion dete hain.

False.

0°

deta hai

q_{0} = cos 0 = 1

;

360°

deta hai

q_{0} = cos 180° = - 1

. Same rotation, opposite quaternion — yahi exactly

q, - q

double cover hai.

Ek rotation matrix ka trace akela rotation angle bata deta hai.

True.

tr (C) = 1 + 2 cos Φ

, isliye

cos Φ = (tr (C) - 1) /2

. Lekin ye axis nahi bataata — uske liye tumhe antisymmetric part chahiye.

Quaternion → Euler → quaternion jaana hamesha original quaternion return karta hai.

False. Euler angles

q

aur

- q

mein fark nahi kar sakte, aur gimbal lock ke paas wo information poori tarah kho dete hain; isliye round trip

- q

ya nearby lekin alag quaternion return kar sakta hai.

Error dhundho

"DCM se Euler angles pane ke liye main

ψ = arctan (C_{12} / C_{11})

use karta hoon."

Seedha

arctan

use karta hai, jo sirf

(- 90°, 90°)

cover karta hai aur dono signs drop kar deta hai —

180°

ki error aati hai jab bhi

C_{11} < 0

ho. Use karo

atan2 (C_{12}, C_{11})

, jo dono signs rakhta hai aur poora circle cover karta hai.

"DCM → quaternion: hamesha

q_{0} = \frac{1}{2} 1 + tr (C)

, phir

q_{1} = (C_{23} - C_{32}) / (4 q_{0})

180°

rotations ke paas toot jaata hai jahan

tr (C) \to - 1

aur

q_{0} \to 0

, to zero se divide ho jaata hai. Shepperd ka fix: pehle

q_{0}, q_{1}, q_{2}, q_{3}

mein se jo sabse bada ho usse compute karo aur usi se divide karo.

"Kyunki

q

aur

- q

same rotation hain,

q

ka sign kabhi bhi kahin matter nahi karta."

Dono same orientation dete hain, lekin attitude interpolation aur error signals ke liye sign matter karta hai: galat sign choose karne se slerp lamba

> 180°

wala path leta hai. Shortest arc ke liye

q_{0} \geq 0

enforce karo.

"Maine quaternion kaafi der tak integrate kiya;

∥ q ∥

1.02

hai lekin theek hai, close enough."

Theek nahi hai. Non-unit

q

se bana

C (q)

ab orthonormal nahi rahega aur vectors ko stretch/shear karega. Har step mein

q \leftarrow q /∥ q ∥

renormalize karo. Dekho Quaternion Kinematics & Propagation.

C = R_{x} (ϕ) R_{y} (θ) R_{z} (ψ)

aur

C = R_{z} (ψ) R_{y} (θ) R_{x} (ϕ)

same rotation hain."

False — rotations commute nahi karte, isliye dono orderings alag DCMs aur alag physical attitudes dete hain. 3-2-1 sequence ka order definition ka hissa hai. Dekho Rotation Sequences (3-2-1 vs 3-1-3).

"Rodrigues mein

[\overset{e}{^}]_{\times}

symmetric hai kyunki ye cross product se aata hai."

Ye antisymmetric hai (

[\overset{e}{^}]_{\times}^{⊤} = - [\overset{e}{^}]_{\times}

); yahi exactly wajah hai ki ye

sin Φ

(odd, direction-flipping) term supply karta hai jabki

I

aur

\overset{e}{^} \overset{e}{^}^{⊤}

even terms supply karte hain. Dekho Rodrigues Rotation Formula.

"Quaternion ko Euler mein convert karne ke liye main ek bada 12-term direct formula derive kar lunga."

Practice ki galti hai, algebra ki nahi: direct formula error-prone hai. DCM ko ek neutral hub ki tarah use karo — quaternion → DCM → Euler — do conversions jo tumhe pehle se trusted hain unhe reuse karo.

"Ek pure

180°

rotation mein

q_{0} = 1

hota hai kyunki ye ek bada turn hai."

Ulta hai.

180°

ke liye

q_{0} = cos (Φ/2) = cos 90° = 0

. Vector part

q

wahaan poori magnitude carry karta hai — yahi exactly wajah hai ki

q_{0}

extraction branch fail hoti hai.

Why questions

Quaternion half angle

Φ/2

kyun use karta hai,

Φ

kyun nahi?

Rotating action

v \mapsto q v q^{- 1}

dono taraf se

q

se multiply karta hai, effectively turn do baar apply karta hai; stored angle ko half karne se wo doubling cancel ho jaati hai to net rotation exactly

Φ

rehta hai.

Angle extraction ke liye

atan2 (y, x)

arctan (y / x)

se better kyun hai?

y / x

divide karne se individual signs chhoot jaate hain, chaar quadrants do mein collapse ho jaate hain.

atan2

dono signs rakhta hai, poora

[- π, π]

range bina ambiguity ke resolve karta hai.

DCM ko conversions ka "hub" kyun kaha jaata hai?

Har representation ka DCM ke saath clean, well-tested formula hota hai, isliye unhe compose karna ek dozen fragile direct formulas yaad karne se behtar hai — sign flip karne ke chances kam hote hain.

Hum quaternion vector part ko

C

ke antisymmetric part se kyun isolate karte hain?

C

ka symmetric part

q_{0}^{2} - q_{1}^{2}

jaisi squares mix karta hai, lekin differences

C_{23} - C_{32} = 4 q_{0} q_{1}

cleanly cross terms pick out karte hain — vector components — diagonal clutter ke bina.

Euler's rotation theorem humein sab kuch axis–angle se derive karne kyun deta hai?

Ye guarantee deta hai ki koi bhi orientation ek axis

\overset{e}{^}

ke baare mein ek single turn

Φ

hai, isliye

(\overset{e}{^}, Φ)

ek complete minimal description hai — DCM aur quaternion sirf uske do repackagings hain. Dekho Euler's Rotation Theorem.

Near-

180°

case specifically trace-based extraction kyun todta hai?

Φ \to 180°

pe,

tr (C) \to - 1

q_{0} = \frac{1}{2} 1 + tr (C) \to 0

; har doosra component phir us vanishing

q_{0}

se divide karke compute hota hai, rounding error catastrophically amplify ho jaata hai.

Gimbal lock tumhe sirf ek degree of freedom kyun cost karta hai, do kyun nahi?

Pitch

θ

ab bhi fully determined hai; sirf yaw aur roll ek single combination

ϕ \pm ψ

mein fuse ho jaate hain. To tum ek independent angle kho dete ho, poora attitude nahi.

C

body-from-inertial hai ya inertial-from-body — ye hamesha state kyun karna chahiye?

Kyunki dono conventions ek doosre ke transposes/inverses hain; unstated convention silently har rotation reverse kar deta hai jo tum apply karte ho aur har attitude jo tum report karte ho.

Edge cases

Identity rotation (

Φ = 0

): uska quaternion aur DCM kya hai?

q = (1, 0, 0, 0)

aur

C = I

. Axis

\overset{e}{^}

undefined hai (kuch turn nahi hota), jo harmless hai — koi bhi axis kaam karta hai kyunki

sin (0) = 0

vector part ko khatam kar deta hai.

x

ke baare mein exactly

180°

rotation, yani

C = diag (1, - 1, - 1)

: kaun sa extraction branch use karte ho?

tr (C) = - 1

isliye

q_{0} = 0

; sabse bade diagonal

C_{11}

pe switch karo, jo deta hai

q_{1} = \frac{1}{2} 1 + C_{11} - C_{22} - C_{33} = 1

, to

q = (0, 1, 0, 0)

Pitch

θ = + 90°

(to

C_{13} = - 1

): yaw aur roll extraction mein kya hota hai?

C_{11} = C_{12} = 0

, isliye

atan2 (0, 0)

undefined hai. Convention se

ϕ = 0

set karo aur determined combination

ϕ + ψ

solve karo; khoya hua DOF report karo. Dekho Gimbal Lock.

Ek degenerate "axis"

\overset{e}{^} = (0, 0, 0)

axis–angle formula ko diya gaya: valid hai?

Nahi —

\overset{e}{^}

ek unit vector hona chahiye. Zero axis ki koi direction nahi hai; sirf identity wale rotation mein koi meaningful axis nahi hoti, use directly

Φ = 0

ke saath represent karo.

Bahut chhota rotation

Φ \approx 0

well-defined axis ke saath: kya kuch numerically fragile hai?

Rotation theek hai, lekin

C

ke tiny antisymmetric part se axis recover karna noisy hai kyunki wo entries

\sim sin Φ \approx Φ

zero ke paas hain — ek small-angle regime jahan quaternions zyada stably integrate karte hain. Dekho Quaternion Kinematics & Propagation.

Thoda non-orthonormal

C

(measurement noise) ko DCM→quaternion mein feed karna: kya toot ta hai?

Trace/off-diagonal identities ab exactly hold nahi karti, isliye

∥ q ∥ \neq = 1

aur

q_{0} = \frac{1}{2} 1 + tr (C)

shayad ek negative number ka square root le le — pehle

C

orthonormalize karo (jaise QUEST-style normalization se).

Same orientation ke liye do alag Euler triples report hue — bug hai ya expected?

Expected. Euler angles periodic aur non-unique hote hain (jaise

ϕ

aur

ϕ \pm 360°

, ya gimbal-lock fusion), isliye distinct triples ek hi attitude naam de sakte hain. DCM ya normalized quaternion canonical arbiter hai.