Cyclic Coordinate Descent Not For The Win

Distinguishing Tentacle Base and Tentacle Origin

#

While implementing CCD IK I realized I need to distinguish between the start of the tentacle (the first segment’s pivot) and the point (0, 0) in the tentacle’s frame of reference. So I’m calling the former the base of the tentacle and the latter the origin. That made it easier to reason about the problem.

CCD IK vs. Simple IK

#

I thought CCD would be better than simple IK. In hindsight, I’m not sure why. I guess because it’s more complicated?

I guess in my earlier tests it did produce results that got closer to the endpoint. However, the results are not stable. They can vary a lot with respect to the initial conditions. Simple IK doesn’t have this problem.

Here’s the player moving around with simple IK. Looks like a fish or something. Nice and streamlined. Each point follows the point before.

Here’s CCD: You can see that the tip of the tentacle doesn’t need to move as much. It looks a little like the fish is dragging something heavy.

This might work for when the tentacle has segments attached to a surface. However, there’s a bigger problem when the tentacle base crosses the tentacle itself. Things shift pretty dramatically in a single frame which is really jarring and un-tentacle-like.

You can see this behaving particularly poorly in this animation: So. I’ll commit CCD IK in Github and then delete it. Might as well remove the debugging output I don’t need while I’m at it.

Meanwhile, here’s even more misbehaviour from CCD IK when the base approaches the tip.

Initial Conditions for IK vs. Direction of IK

#

A realization. Yesterday I said, “I was conflating the direction that the IK algorithm is applied and the initial positions of the segments. I can try doing those separately.” That’s not true at all. I can’t do those separately.

If I use the initial conditions where the tentacle base is at the origin and then try to iteratively move the base toward the origin… you see the problem.

If X is the tentacle base and O is the point where the tentacle is attached to the surface then the diagram on the left means IK from base to tip and the one on the right is tip to base. The idea of using initial conditions that are a blend of these two states is interesting. The tentacle lengths would not be constant so unless I have some way of making them gradually shift from their initial blended values to their actual values I think it’d look pretty wonky.

Could be interesting to experiment with but I’ll probably skip it for greater causes.

Still worth investigating placing the tentacle base at the origin, doing IK to close the gap to where the tentacle is attached to the surface, then updating the tentacle segment points to move the error from the surface to the base.

The alternative to that is to maintain some rigidity in the tentacle, especially near the base. That seems like a more intuitive solution. Downside: it involves adding another layer of complexity.

Maybe leave it as is for now. I want to work on getting the tentacle to explore forward in the direction of the player movement, attach to something, then push the player forward.

Manual Tentacle Walk - Now With Widescreen!

#

I widened the sketch window and put the starting position a little above the ground and closer to the left wall, the better for prototyping the walk.

There are three phases. Reach forward, reach toward the ground, and then push the player forward. I want to overlap the first two phases in a natural way.

I also realized it should only attach to surfaces it collides with during the second and maybe third phase. During non-attach phases maybe it can pull back from any surfaces it comes in contact with. I guess all nearby segments could contract to varying degrees to facilitate that.

Not sure but I might need to give the tentacle some thickness for that. So it can detect which side to pull away from.