Re: [ggt-devel] Question about EulerAngleZXY, XYZ, ZYX

[Doug M. <mc...@ia...>]

Doug McCorkle wrote:
> Allen Bierbaum wrote:
> 
>> Doug McCorkle wrote:
>>
>>> Allen Bierbaum wrote:
>>>
>>>> Doug McCorkle wrote:
>>>>
>>>>> Allen Bierbaum wrote:
>>>>>
>>>>>> Doug McCorkle wrote:
>>>>>>
>>>>>>> Allen Bierbaum wrote:
>>>>>>>
>>>>>>>> Allen Bierbaum wrote:
>>>>>>>>
>>>>>>>>> Unfortunately, it looks like there are some bugs in the 
>>>>>>>>> conversion somewhere.  My guess is that there are some border 
>>>>>>>>> cases in the extraction of the euler angles from the Matrix.
>>>>>>>>>
>>>>>>>>> I have added a new test case that exhaustively tests all Euler 
>>>>>>>>> getting and setting to find these issues. The test can be found 
>>>>>>>>> in the MatrixGenTest test case.  If anyone has some time to 
>>>>>>>>> look into this, you can use this test case to reproduce the 
>>>>>>>>> problem.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> I just commited some changes that should fix the errors. (at 
>>>>>>>> least it fixes the ones the test case found :)
>>>>>>>>
>>>>>>>> Give it a spin and let me know.
>>>>>>>>
>>>>>>> Just do a cvs checkout I assume?
>>>>>>>
>>>>>> Yep.
>>>>>>
>>>>>>> Thanks for the prompt response. This was going to be my evening 
>>>>>>> project but I guess I can do some testing instead. I appreciate 
>>>>>>> the help.
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> I wasn't planning on working on it, but I had some down time 
>>>>>> waiting for some builds and I found a good solid algorithm in a 
>>>>>> reference so I just gave it a try.
>>>>>>
>>>>> Ok. It appears that things are better but not quite right. If 
>>>>> rotations are purely about an individual axis X,Y,Z everything 
>>>>> works fine. If rotations become compounded things begin to act odd. 
>>>>> This affect may be due to us not understanding how rotations are 
>>>>> applied in OSG though so I would like to do more testing. Some 
>>>>> things do appear to working better but one odd thing did occur. 
>>>>> With EulerXYZ under the old version of gmtl  the compound rotations 
>>>>> appeared to work correctly no mater what had been done. This does 
>>>>> not appear to be working now. I will continue to test and try to 
>>>>> provide better feedback.
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> One point of feedback.  Because Euler angles are non-unique, 
>>>> sometimes the angles returned do not appear to be continuous.  There 
>>>> is really no way around this.  The test code I wrote sets a matrix 
>>>> using a set of euler angles and then extracts the euler angles and 
>>>> uses the extracted eulers to make a new matrix.  Then it tests if 
>>>> the two matrices are equal.  Note, that this does not require that 
>>>> the original euler values and the extracted euler values must 
>>>> match.  It only requires that the values are functionally 
>>>> equivalent.  The test shows that this works for all angles in the 
>>>> range of -190 degrees to +190 degrees for x,y, and z in all 3 euler 
>>>> types.
>>>>
>>> That is what I am seeing. The angles are not continuous. Is there a 
>>> better way to handle this? I need to extract rotations for XYZ out of 
>>> a matrix. I thought that EulerAngle would be the best way. Is there a 
>>> better way?
>>
>>
>>
>> Depends upon why you need them to be continuous.  You could always use 
>> Quaternions if you needs something that has numbers that look more 
>> continuous.
>>
> Well, functionally should be ok it just doesn't appear that that is what 
> is happening. I think I may be messing up how OSG is expecting the 
> angles to be applied to the transform. I need to do more testing. Thanks 
> again for the help.
> 
> Doug
> 

I went ahead and implemented quats and everything works much smoother. I 
am not sure why we initially implemented Euler angles. Thanks for all 
the help. I suppose there is a least one less bug in gmtl though.

Doug

>> -Allen
>>
>>>
>>>> -Allen
>>>>
>>>>>
>>>>>> -A
>>>>>>
>>>>>>>
>>>>>>>> -A
>>>>>>>>
>>>>>>>>>
>>>>>>>>> -Allen
>>>>>>>>>
>>>>>>>>> Doug McCorkle wrote:
>>>>>>>>>
>>>>>>>>>>  Hello,
>>>>>>>>>>      When using gtml I noticed that when I pass an identity 
>>>>>>>>>> Matrix44f into
>>>>>>>>>>  make< gmtl::EulerAngleZXY >( identity ) that I get back 
>>>>>>>>>> 90,0,90. If I
>>>>>>>>>>  ask for an EulerAngleXYZ I get back 0,0,0 and ZYX return 0, 
>>>>>>>>>> nan, 0. Is
>>>>>>>>>>  this the correct behavior of EulerAngle? I thought that no 
>>>>>>>>>> matter what
>>>>>>>>>>  Euler combination that I ask for should return 0,0,0 for an 
>>>>>>>>>> identity
>>>>>>>>>>  matrix. Thanks for the help in understanding this better. I 
>>>>>>>>>> am using
>>>>>>>>>>  gmtl 0.4.5.
>>>>>>>>>>
>>>>>>>>>>
>>>