SetIKHintPosition and SetIKHintPositionWeight seems to have no impact

Question

I have recently been exploring Unity's inverse kinematics, to further polish the interaction between our rigged avatars and props.

I have made a sample project, using PuppetMaster's basic humanoid for the model, rig, and idle animation, but without Final IK; meaning this is Unity's inverse kinematics. Inside our OnAnimatorIK method, we set the right elbow's SetIKHintPosition and SetIKHintPositionWeight like we did with the hand IK:

animator.SetIKHintPosition(avatarIKHint, hintPosition);
animator.SetIKHintPositionWeight(avatarIKHint, hintBlendValue);

animator.SetIKPosition(sourceIKGoal, targetPosition);
animator.SetIKPositionWeight(sourceIKGoal, blendValue);

You can see the left and right hand correctly move to where we set it. However, it seems like the right elbow is ignoring the hint position entirely. You can see the white debug line connecting the elbow to the hint position in the picture above.

It seems like the elbow ONLY bends if it absolutely must for the hand to reach the target position, regardless of where the hint is set, and the weight value given.

I have fiddled with this on and off for days and still haven't found a solution. Others have asked the same question here and on other sites, but no one has posted an answer.

Answer 1

I notice that the AvatarIKHint enumeration supports applying hints to only the elbows and knees.

This suggests to me that Unity uses a simple 2-bone IK solution for this case. That means it thinks of the shoulder or hip joint as fixed, and finds the best solution it can using only the upper arm & forearm bones, or thigh and shin bones for legs.

The advantage of this approach is it's fast: you can find the best solution in a single step with a little high school trigonometry, without having to iterate over longer bone chains and converge gradually toward a solution.

Now, a 2-bone IK system with a root point (shoulder / hip) and a target point has one extra degree of freedom: the rotation about the line from the root to the target. Imagine doing the arm-flapping motion of the chicken dance: the line from your shoulder to your wrist stays fixed, as does the joint angle at your elbow; all that changes is the rotation at your shoulder to position your elbow low/tucked-in close to your body or raised outward. All of these poses are potential, equally-valid solutions for this IK problem, so if we care which of these poses we adopt, then we need to hint the solver in the desired direction.

It looks like that degree of freedom is what these hint methods serve to control. So for instance if your FPS character was putting their elbow way out to the side like a novice when holding their weapon, you could use a hint to guide the elbow down to a more professional-looking tucked-in position.

Once the arm or leg is fully extended however, there's no extra degree of freedom left: the elbow/knee has to sit along the line from the shoulder/hip to the wrist/ankle. Putting the elbow anywhere else would make the end effector deviate from the target (bad) or distort the skeleton (also bad).

So, it looks like what you're trying to accomplish with this hint is not what this hint is made to do. What you should do instead depends on what your ultimate aim is.

If you want the character to lean to the side, bringing the shoulder closer to give the elbow more room to bend & position itself away from the direct shoulder-to-target line, then you might have luck applying that lean as an additive animation layer.