procedural step logic

[Novakasa]

1. find the foot that is more likely to get moved first
2. determine whether that foot needs to move, now the predicted step size can be computed relative to the other foot, since we can assume that the other foot will not move first

left/right offset should be proportional to how long we expect the foot to be placed there, since it requires more stability

[Novakasa]

how much of the foot cycle is spent locked should be proportional to spring force. When spring force is high compared to just standing normal force, most of the time one foot should be locked (half a cycle for each foot)

[Novakasa]

probably a better decision when to unlock foot:
length_to_hip > some threshold, and the change to this length is positive.

We also need a way to prevent "hover feet" where a foot decides not to lock after a while, this probably happens when the "unlock candidate" changes frequently between the two feet

[Novakasa]

probably have some heuristic on what the actual distance to predicted target is, which then influences how fast we have to move there

[Novakasa]

we should probably have a procedural hip that bounces depending on locked feet? or a seperate center of mass

[Novakasa]

probably proceed with visual center of mass that is allowed to oscillate based on locked feet and applied forces

regarding the unlock time, we need to add an upper limit for how long no contact is made and probably a lower limit to how short a contact should be

[Novakasa]

these limits could be set dynamically by simulating the center of mass "bounce", combining it with rig center of mass. While no contact is made, center of mass is in free fall, and this should be compensated during feet contact. So we need to make sure that foot is not released while there is negative up velocity on simulated center of mass, or a similar criterion