Closed gabrielenava closed 8 years ago
The backslash-operator is better and faster than the ^{-1} operation. Use it as often if it is possible!! Because this operator can be very well parallelized and also supports gpuArray
which runs on GPU!
You solve there a linear equation with -Ax = B, s.t. x = -A \ B and then you divide x with A.
There exists a really good book for Matalb-Hacks: Yair Altman, Accelerating MATLAB ® Performance, ISBN: 978-1-4822-1130-6 (eBook - PDF)
See also:
@Ganimed I agree, but is not exactly readable for not matlab experts, for that I was suggesting to add a comment to explain what was going on.
I do not know the complete code there. I'm not 100%ly sure but does this line describe a similarity transformation?
@Ganimed no, it is implementing the computation of the derivative of a inverse matrix, as in https://en.wikipedia.org/wiki/Invertible_matrix#Derivative_of_the_matrix_inverse .
Ok, I understand now. Thanks for the info.
I would suggest that you explain in your comment exactly that what you have explained to me (incl. the link to wikipedia).
@gabrielenava please revisit notation of transformations. For instance, if a transformation T
expresses a vector v_a
in a coordinate system A
into the coordinate system B
, denote it with
b_T_a
Maybe it is better to use an other notation instead of b_T_a
. We all know what this means and also how it is expressed in the books. But not everybody knows that. And also on the screen the human is intuitively used to read from left to right (except in the arabic world). Maybe the notation, T_a2b
, is better for that and easier to read instead of b_T_a
.
@Ganimed the nice thing about having the orientation of the right and the reference orientation on the left is that than is to check that the composition of rotation make sense, by simply comparing nearby symbols , compare:
a_T_d = a_T_b * b_T_c * c_T_d
with
T_d2a = T_b2a * T_c2b * T_d2c
I can easily say that if the first one is right or wrong, while is difficult to do this kind of visual check on the second expression.
I vote for the notation of a_T_b. It is a lot lot easier to debug.
On Mon, Nov 9, 2015 at 12:54 PM, Silvio Traversaro <notifications@github.com
wrote:
@Ganimed https://github.com/Ganimed the nice thing about having the orientation of the right and the reference orientation on the left is that than is to check that the composition of rotation make sense, by simply comparing nearby symbols , compare:
a_T_d = a_T_b * b_T_c * c_T_d
with
T_d2a = T_b2a * T_c2b * T_d2c
I can easily say that if the first one is right or wrong, while is difficult to do this kind of visual check on the second expression.
— Reply to this email directly or view it on GitHub https://github.com/robotology/mex-wholebodymodel/pull/57#issuecomment-155044181 .
Ok, I understand, I got your points.
:+1:
Last commit addresses https://github.com/robotology/mex-wholebodymodel/issues/56
I think we can merge then.
By the way, from now on I will use this notation-style also for my code.
:+1:
@gabrielenava @naveenoid are you ok with merging?
:+1:
both the original and joint space controllers have been updated to the last version. they both passed the balancing tests. README files updated