THESIS WORK: Footprint-Based Quadruped Motion Synthesis
by Nick Torkos
Keywords: Footprints, Trajectory Optimization, articulated figure animation, physically-based animation

(The Full Postscript Thesis can be downloaded here)


Both people and legged animals leave trails of footprints on the ground when walking along an impressionable surface. As such, we all have some understanding of how footprints are related to the motions that create them. A computational method that can generate realistic animations from footprints is therefore potentially a powerful animation tool for use by animators of all levels of experience.

How can the motion be reconstructed from a set of footprints? The main difficulty in answering this question arises from the fact that footprints only constrain the feet of the animal. To create an animation, all of the unconstrained body parts must be animated in a way that is both physically realistic and visually pleasing.

The use of physics is essential in the process of motion reconstruction because footprint positions and times might necessitate ballistic motions. However, the amount of physical simulation required is debateable. Full physical simulations of articulated animals is of O(n) complexity for each time step but is in practice a computationally expensive proposition.

Though full simulations take care of the motion of every joint, they introduce the harder problem of control. Any movement in one limb has repercussions in the motion of the other limbs so great care must be taken in coordinating the amount of force that the muscles exert on each joint. This has limited the usage of true physical simulations to only simple models.

Our approach uses simple physics applied to an underlying approximation of the original animal, as opposed to fully simulating the physics over the entire skeleton. Not only does this simplify the entire procedure of generating the animation, but it also makes it easier to guide the motion generation with what we call comfort constraints, in order to control the quality of the motion. The motion of the underlying approximation contains a lof of data that can then be used to guide the kinematic constraints of the original animal model so that the final animation appears realistic. Specific knowledge of the animal's structure is also used to assist in the motion reconstruction. In all cases, we purposefully use methods that can be described quantitatively as well as having a readily accessible qualitative interpretation in order to maintain the intuitive feel of the animation tool.