converting quaternion to euler angles | Together C & C++ | Page 1

rapid swallow Dec 10, 2022, 6:23 AM

#

I am trying to convert a quaternion to a 2 dimensional rotation vector. i have successfully converted the 2d rotation vector to quaternions but going backwards is giving me a huge headache. im wondering if anyone knew how exactly to string together the converting back properly.

dapper kettleBOT Dec 10, 2022, 6:23 AM

#

When your question is answered use !solved to mark the question as resolved.

Remember to ask specific questions, provide necessary details, and reduce your question to its simplest form. For tips on how to ask a good question run !howto ask.

rapid swallow Dec 10, 2022, 6:25 AM

#

I expect quaternionToEulerRot to give me -70.f, 125.f

#

I heavily based my formulas on this article: https://en.wikipedia.org/wiki/Conversion_between_quaternions_and_Euler_angles, had to change some stuff around to get the expected result for eulerRotToQuaternion but applying the changes to the reverse formula is just not making sense to me

#

converting quaternion to euler angles

#

📎 main.cpp

jovial verge Dec 10, 2022, 7:12 AM

#

your y and z (and their signs) are flipped in line 40-41

#

additionally, on line 57 the x/z and w/y should also be switched

#

line 40/41:

(cx * sy * cz) - (sx * cy * sz), // quat.y
(sx * cy * cz) + (cx * sy * sz)  // quat.z

line 57:

float sinp = 2.f * (quat.x * quat.z - quat.w * quat.y);

#

btw, (after the 40/41 fix) your w and x are swapped compared to wikipedia and likewise for z and y
this is why they need to be switched for line 57 (the other ones are all addition)

#

@rapid swallow

rapid swallow Dec 10, 2022, 7:19 AM

#

jovial verge <@628510537138962432>

I actually intentionally switched some stuff around for eulerRotToQuaternion, and that formula produces expected results

#

i think my target has roll and pitch flipped or something

#

or something flipped not sure, but i just know that eulerRotToQuaternion is "correct"*

jovial verge Dec 10, 2022, 7:21 AM

#

oh ok, then i guess you just need to switch quat.y and quat.z in your quaternionToEulerRot and flip the signs for both

rapid swallow Dec 10, 2022, 7:23 AM

#

wym "flip the signs for both"?

#

like, the members of the final result?

jovial verge Dec 10, 2022, 7:23 AM

#

change (sx * cy * cz) - (cx * sy * sz) to +

#

and same for z

rapid swallow Dec 10, 2022, 7:25 AM

#

But, as I said above, eulerRotToQuaternion gives me correct numbers

#

dont really know why but it does lol

jovial verge Dec 10, 2022, 7:25 AM

#

huh...

rapid swallow Dec 10, 2022, 7:25 AM

#

the relationship is really close as you can see

#

swap the members and take their complement or something...but thats cheeky

jovial verge Dec 10, 2022, 7:26 AM

#

wait you mean the r1 numbers? or the q2?

rapid swallow Dec 10, 2022, 7:27 AM

#

r1 is what im starting with

#

in my console i expect to be able to convert r1 to a quat then back to a vec2 and end up with the same values that i started with

jovial verge Dec 10, 2022, 7:27 AM

#

jovial verge line 40/41: ```c++ (cx * sy * cz) - (sx * cy * sz), // quat.y (sx * cy * cz) + (...

well these 2 changes fix r1

#

(the second one fixes sign of first number, and the first one fixes the 55/110 to 70/125)

rapid swallow Dec 10, 2022, 7:31 AM

#

one moment...

#

i think its working now

#

tysm!!

jovial verge Dec 10, 2022, 7:38 AM

#

oh um....

#

it seems that rot.x has to be within (-90, 90)

#

probably because of the asin

rapid swallow Dec 10, 2022, 7:38 AM

#

jovial verge well these 2 changes fix r1

I applied your requested changes here

rapid swallow Dec 10, 2022, 7:38 AM

#

jovial verge it seems that rot.x has to be within (-90, 90)

it just so happens that those are the bounds of my rot

#

its for a player

#

if they look straight up, they are at -89.99 degrees

#

straight down, 89.99

#

should i be concerned with that?

jovial verge Dec 10, 2022, 7:39 AM

#

uh gimme 1 sec

rapid swallow Dec 10, 2022, 7:40 AM

#

rapid swallow if they look straight up, they are at -89.99 degrees

so the player's head x rot is in a range of (-90, 90), head y rot is (-180, 180)

#

current code fwiw

📎 main.cpp

jovial verge Dec 10, 2022, 7:41 AM

#

gah i hate floating points

#

would it be possible to switch out the float for doubles?

#

sometimes it just likes to break on me and i don't exactly know why

rapid swallow Dec 10, 2022, 7:44 AM

#

jovial verge would it be possible to switch out the float for doubles?

sure

#

I mean, my target wants floats but its the same dif

jovial verge Dec 10, 2022, 7:45 AM

#

are you getting floats as inputs as well?

rapid swallow Dec 10, 2022, 7:45 AM

#

yes

jovial verge Dec 10, 2022, 7:45 AM

#

mm ok

rapid swallow Dec 10, 2022, 7:45 AM

#

the quat is just gonna be like that vec4 thing with 4 floats

jovial verge Dec 10, 2022, 7:46 AM

#

it would probably be better if you used doubles for the processing stuffs, to minimize error

rapid swallow Dec 10, 2022, 7:47 AM

#

yes, I see

jovial verge Dec 10, 2022, 7:47 AM

#

i'm getting some pretty large errors here lmao :\ (>0.05)

rapid swallow Dec 10, 2022, 7:47 AM

#

just by virtue of floating point inaccuracies?

#

i should prob know this lol but do the errors "build up" over time?

jovial verge Dec 10, 2022, 7:48 AM

#

i think the errors happen more toward the extremes

rapid swallow Dec 10, 2022, 7:48 AM

#

like, would i benefit if i were to do everything in terms of doubles then cast the final result to a float?

jovial verge Dec 10, 2022, 7:48 AM

#

probably... but i can check

rapid swallow Dec 10, 2022, 7:49 AM

#

I can't change the actual inputs and outputs to doubles because im working with someone elses api

jovial verge Dec 10, 2022, 7:50 AM

#

understandable
a lot of graphics stuff is done in float becuse i think even modern gpus sometimes don't support double lol

rapid swallow Dec 10, 2022, 7:50 AM

#

yea i see

#

but i totally can use doubles as an intermediary step

jovial verge Dec 10, 2022, 7:51 AM

#

i can't tell if it's better or the same lmao

#

i'm getting 89.9984, -166.899 -> 90, -166.854

#

89.9802, -169.836 with float (the first one was double as intermediary)

rapid swallow Dec 10, 2022, 7:53 AM

#

i guess only way to tell is if you actually test a double input and output then compare that

jovial verge Dec 10, 2022, 7:53 AM

#

seems like a negligible difference to me

rapid swallow Dec 10, 2022, 7:53 AM

#

i suppose so

#

but in practice i should keep in mind how lots of repeated arithmetic might have noticeable precision errors with floats

jovial verge Dec 10, 2022, 7:54 AM

#

here is a random "fuzz" thing i hacked together (it is not very good code)

Vec2 r2;
    for (float i = -89.9f; i < 90.0f; i += 0.1f)
    {
        for (float j = -179.9f; j < 180.0f; j += 0.1f)
        {
            r2.x = i;
            r2.y = j;
            auto q2 = eulerRotToQuaternion(r2);
            auto r1 = quaternionToEulerRot(q2);
            if (std::abs(r1.x - r2.x) >= 0.1 || std::abs(r1.y - r2.y) >= 0.1)
            {
                std::cout << i << " " << j << std::endl <<
                    r1.x << " " << r2.x << "\t" << r1.y << " " << r2.y << std::endl;
            }
        }
    }

#

i unfortunately have to leave, but you can play around with the things and see if the errors are acceptable

#

or whatever, idk

rapid swallow Dec 10, 2022, 7:56 AM

#

thats ok thx again for ur help

#

prob wouldnt have figured this out otherwise lol

jovial verge Dec 10, 2022, 7:57 AM

#

@rapid swallow Has your question been resolved? If so, run !solved :)

rapid swallow Dec 10, 2022, 7:59 AM

#

oh yes

#

soryyy

#

!solved

#converting quaternion to euler angles