Very nice! How much "setup by artist" does it need? I guess that you are trying to make it as generic as possible to be out of the box? I also want to get down to implement it one day.

The elephant seems to have very low traction with the ground, almost as if its walking on ice. Do you have a method to deal with that?

Both the elephant and the croc (but more so the elephant) seem to be sliding a little - the leg movement isn't looking too bad otherwise though!

This is tricky stuff to get right, lots of fiddly little adjustments! 

I'm glad you guys were interested, my implementation is very comically bad at the moment , but in the game the models are so small on screen I doubt players will notice. If they were relatively bigger I'd spend a lot longer on getting this right, in fact I will probably come back to it, but I feel I've wasted a couple of weeks on tweaking this already!

The sliding isn't actually as bad as in these videos, I should have said, in both cases I had scaled up the models just for the videos so the animation speed doesn't match the movement speed. In the game the match is much better as the animation speed is driven off how far the animals moves, I will try to do a later video soon with all in game.

Also the base animation for the croc is a very bad test anyway, I basically just wiggled the legs in blender!  Obviously also these are low poly and the skin weightings need some work too, I didn't envisage the joints moving so far when I made the models.

I have seen there is middleware available for doing this with quadrupeds, IKinema and Bik. I suspect they use a better iterative method and I will try this if I have time. But even with iterative I'm sure it took a lot of tweaking to get right. And I am slightly worried it will use more performance as it has to work on mobiles.

I could write much more on how it works (or fails to work lol) if people were interested.

One major change I would make if possible is instead of calculating the solution each frame from the original animation, I would calculate the solution from the previous frame solution, and limit the change in rotations (or even use damping), to prevent physically incorrect snapping. Also as a performance optimization and to further smooth it might be an idea to calculate the solution at a fixed lower tick rate (say 30fps or matching the underlying animation tick rate) and interpolate frames.

Here is example of setup for the most complex animal, crocodile. The others are simpler, especially 2 legged. All this is very hacky and subject to change but should give an idea that setup is not very involved, most is reading bone IDs from a txt file. I'll probably replace it with something much simpler.

num_feet = 4;
// Leg IK chains, end bone, chain length, chain effector, knee joint flip, shoulder joint flip
chain_leg[0].Create(23, 3, 0, true, false);
chain_leg[1].Create(27, 3, 0, true, false);
chain_leg[2].Create(12, 3, 0, false, true);
chain_leg[3].Create(16, 3, 0, false, true);

// Limit knee angles (smallest, largest)
ANGLES_FRONT(0, 104);
ANGLES_BACK(0, 80);

// hips and shoulders, might be inferred from the chains later
SetLegRoots(22, 26, 11, 15);

// chain for back, end bone ID, num bones, effector
chain_back.Create(2, 2, 1);
// chain for tail
chain_tail.Create(19, 3, 0);

// offset in model space from last tail bone joint
m_TailOffset_MS = 6000;
// croc uses an extra 3rd probe for torso orientation
Set_UseMiddleProbe(true);
// neck and 'nose' bone
SetHead(3, 5);

// used for calcing head angles, some of these might be inferred from bones
m_NoseOffset_MS = 4000;
m_NeckNoseDist_WS = 0.2f * Scale;
m_DesiredHeadHeight_WS = 0.1f * Scale;
Set_TiltHeadDown(true); // tilt head down when going down slopes
m_GlobalZOffset_MS = 1500.0f; // overall bodge to the torso height above ground