I remember the first time I tried to set up a seamless IK FK switch with Python vividly. There was this mechanical EVA suit that I was rigging for a masterclass assignment at uni given by Frontier. The IK to FK switching was trivial and there were not many issues with that, but I had a very hard time figuring out the FK to IK one, as I had no idea what the pole vector really is and also, my IK control was not oriented the same way as my FK one.
Im sure that throughout the web there are many solutions to the problem, but most of the ones I found were in MEL and some of them were a bit unstable, because they were relying too much on the
xform command or the
rotate one with the
ws flag, which I am assuming causes issues sometimes when mapping from world space to relative, where a joint will have the exact same world rotation, so it looks perfectly, but if you blend between IK and FK you can see it shifting and then coming back in place. That’s why I decided to use constraints to achieve my rotations, which seems to be a simple enough and stable solution.
EDIT: It seems like even with constraints it is possible to get that issue in the case where the IK control is oriented differently. What fixes is though is switching back and forth once more.
Here is what we are trying to achieve
Basically, there is just one command for the seamless IK FK Switch, which detects the current kinematics and switches to the other one maintaining the pose. I have added the button to a custom marking menu for easier access.
So, in order to give you a bit of a better context I have uploaded the example scene that I am using, so you can have a look at the exact structure, but feel free to use your own scene with IK/FK blending setup. The full code (which is very short anyway) is in this gist and there are three scene files in here for each version of our setup. The files contain just a simple IK/FK blending system, on which we can test our matching setup, but with different control orientations.
It is important to understand the limitations of a seamless IK FK switch before we dive in. Mainly, I am talking about the limited rotation of the second joint in the chain, as IK setups allow for rotations only in one axis. What this means is that if we have rotations in multiple axis on our FK control for that middle joint (elbow, knee, etc.) the IK/FK matching will not work properly. All this is due to the nature of inverse kinematics.
Also, for easier explaining I assume we are working on an arm and hand setup, but obviously the same approach would work for any IK/FK chain.
We will consider three cases:
– All controls and joints are oriented the same
– IK Control oriented in world space
– IK Control and IK hand joint both oriented in world
Again, you do not have to use the same file as I do as it is just an example, but it is important to be clear on the existing setup. We assume that we have an arm joint chain – L_arm01_JNT > L_arm02_JNT > L_arm03_JNT and a hand joint chain – L_hand01_JNT > L_hand02_JNT with their correspondent IK and FK chains – L_armIk01_JNT > …, L_armFk01_JNT > …, etc. These two chains are blended via a few
blendColors nodes for translate, rotate and scale into the final chain. The blending is controlled by L_armIkFk_CTL.fkIk. Then we have a simple non-stretchy IK setup, but obviously a stretchy one would work in the same way. Lastly, the
L_hand01_JNT is point constrained to
L_arm03_JNT and we only blend the
scale attributes on it, as otherwise the wrist becomes dislocated during blending, because we are interpolating linearly translation values.
Now that we know what we have to work with, let us get on with it.
So, in this case, all of our controls and joints have the exact same orientation in both IK and FK. What this means is that essentially all we need to do to match the kinematics is to just plug the rotations from one setup to the other. Let’s have a look. The scene file for this one is called ikFkSwitch_sameOrient.ma
IK to FK
This one is always the easier setup, as FK controls generally just need to get the same rotation values as the IK joints and that’s it. Now, initially I tried copying the rotation via
xform commands, but whenever a control was rotated a bit too extreme these would cause flipping when blending between IK and FK, which I am assuming is because these commands have a hard time converting the world space rotation to a relative one, causing differences of 360 degrees. So, even though in full FK and full IK everything looks perfect, in-between the joint rotates 360 degrees. Luckily, maya has provided us with constraints which have all the math complexity built in. Assuming you have named your joints the same way as me we use the following code.
mc.delete(mc.orientConstraint("L_armIk01_JNT", "L_armFk01_CTL")) mc.delete(mc.orientConstraint("L_armIk02_JNT", "L_armFk02_CTL")) mc.delete(mc.orientConstraint("L_handIk01_JNT", "L_handFk01_CTL")) mc.setAttr("L_armIkFk_CTL.fkIk", 0)
As I said, this one is fairly trivial. We just orient each of our FK controls to match the rotations of the IK joints. Then in the end we change our blending control to FK to finalize the switch.
FK to IK
Now, this one was a pain the first time I was trying to do it, because I had no idea how pole vectors worked at all. As soon as I understood that all we need to know about them is that they just need to lie on the same plane as the three joints in the chain, it became easy. So essentially, we need to place the IK control on the FK joint to solve the end position. And then to get the elbow (or whatever your mid joint is representing) to match the FK, we just place the pole vector control at the exact location of the corresponding joint in the FK chain. So, we get something like this.
mc.delete(mc.parentConstraint("L_handFk01_JNT", "L_armIk_CTL")) mc.xform("L_armPv_CTL", t=mc.xform("L_armFk02_JNT", t=1, q=1, ws=1), ws=1) mc.setAttr("L_armIkFk_CTL.fkIk", 1)
Now even though this does the matching job perfectly it is not great for the animators to have the control snap at the mid joint location as it might go inside the geometry, which is just an unnecessary pain. What we can do is, get the two vectors from arm01 to arm02 and from arm03 to arm02 and use them to offset our pole vector a bit. Here’s the way we do that.
arm01Vec = [mc.xform("L_armFk02_JNT", t=1, ws=1, q=1)[i] - mc.xform("L_armFk01_JNT", t=1, ws=1, q=1)[i] for i in range(3)] arm02Vec = [mc.xform("L_armFk02_JNT", t=1, ws=1, q=1)[i] - mc.xform("L_armFk03_JNT", t=1, ws=1, q=1)[i] for i in range(3)] mc.xform("L_armPv_CTL", t=[mc.xform("L_armFk02_JNT", t=1, q=1, ws=1)[i] + arm01Vec[i] * .75 + arm02Vec[i] * .75 for i in range(3)], ws=1)
xform returns lists in order to subtract them to get the vectors we just loop through them and subtract the individual elements. If you are new to the shorter loop form in Python have a look at this. Then once we have the two vectors we add 75% of them to the position of the
arm02 FK joint and we arrive at a position slightly offset from the elbow, but still on the same plane, thus the matching is still precise. Then our whole FK to IK code would look like this
mc.delete(mc.parentConstraint("L_handFk01_JNT", "L_armIk_CTL")) arm01Vec = [mc.xform("L_armFk02_JNT", t=1, ws=1, q=1)[i] - mc.xform("L_armFk01_JNT", t=1, ws=1, q=1)[i] for i in range(3)] arm02Vec = [mc.xform("L_armFk02_JNT", t=1, ws=1, q=1)[i] - mc.xform("L_armFk03_JNT", t=1, ws=1, q=1)[i] for i in range(3)] mc.xform("L_armPv_CTL", t=[mc.xform("L_armFk02_JNT", t=1, q=1, ws=1)[i] + arm01Vec[i] * .75 + arm02Vec[i] * .75 for i in range(3)], ws=1) mc.setAttr("L_armIkFk_CTL.fkIk", 1)
Now, in this case, the orientation of the IK control is not the same as the
hand01 joint. I think in most cases people go for that kind of setup as it is much nicer for animators to have the world axis to work with in IK. The scene file for this one is called ikFkSwitch_ikWorld.ma.
The IK to FK switch is exactly the same as the previous one, so we will skip it.
FK to IK
So, in order to get this to work, we need to do the same as what we did in the previous case, but introduce an offset for our IK control. How do we get this offset then? Well, since we can apply transformations only on the controls, we need to calculate what rotation we need to apply to that control in order to get the desired rotation. Even though, we can calculate the offsets using maths and then apply them using maths, we might run into the same issue with flipping that I discussed in the previous case. So, instead, a much easier solution, but somewhat dirtier is to create a
locator which will act as our dummy object to orient to.
Then, in our case where only the IK control is oriented differently from the joints, what we need to do is create a
locator and have it assume the transformation of the IK control. The easiest way would be to just parent it underneath the control and zero out the transformations. Then parent the
locator to the
L_handFk01_JNT, as that’s the one that we want to match to. Now wherever that handFk01 joint goes, we have the
locator parented underneath which shares the same orientation as our IK control. Therefore, just using
parentConstraint will give us our matching pose. Assuming the locator is called L_hand01IkOfs_LOC all we do is this.
This will get our wrist match the pose perfectly. Then we apply the same code as before to get the pole vector to match as well and set the IK/FK blend attribute to IK.
arm01Vec = [mc.xform("L_armFk02_JNT", t=1, ws=1, q=1)[i] - mc.xform("L_armFk01_JNT", t=1, ws=1, q=1)[i] for i in range(3)] arm02Vec = [mc.xform("L_armFk02_JNT", t=1, ws=1, q=1)[i] - mc.xform("L_armFk03_JNT", t=1, ws=1, q=1)[i] for i in range(3)] mc.xform("L_armPv_CTL", t=[mc.xform("L_armFk02_JNT", t=1, q=1, ws=1)[i] + arm01Vec[i] * .75 + arm02Vec[i] * .75 for i in range(3)], ws=1) mc.setAttr("L_armIkFk_CTL.fkIk", 1)
Now, in this last scenario, we have the
handIk01 joint oriented in world space, as well as the control. The reason you would want to do this again is to give the animators the easiest way to interact with the hand. In the previous case, the axis of the IK control do not properly align with the joint which is a bit awkward. So a solution would be to have the
handIk01 joint oriented in the same space as our control, so the rotation is 1 to 1 and it is a bit more intuitive. The scene for this one is
ikFkSwitch_ikJointWorld.ma and it looks like this.
It is important to note that the IK joint is just rotated to match the position of the control, but the
jointOrient attributes are still the same as the FK one and the blend one.
So again, going from IK to FK is the same as before, we are skipping it. Let us have a look at the FK to IK.
FK to IK
This one is very similar to the previous one, where we have an offset transform object to snap to. The difference is that now instead of having that offset be calculated just from the difference between the IK control and the FK joint, we also need to adjust for the existing rotation of the IK joint as well. So, we start with our
locator the same way as before – parent it to the IK control, zero out transformations and parent to the
handFk01 joint. And then, the extra step here is to apply the negative rotation of the IK joint to the
locator in order to get the needed offset. So, this calculation would look like this.
ikRot = [-1 * mc.xform("L_handIk01_JNT", ro=1 ,q=1)[i] for i in range(3)] mc.xform("L_hand01IkOfs_LOC", ro=ikRot, r=1)
We just take the rotation of the IK joint and multiply it by -1, which we then apply as a relative rotation to the
And then again, as previously we just apply the pole vector calculation and we’re done.
So, as you can see, scripting a seamless IK FK switch is not really that complicated at all, but if you are trying to figure it out for the first time, without being very familiar with rigging and 3D maths it might be a bit of a pain. Again, if you want to see the full code it is in this gist.