General mechanisms for keyframe selection and motion editing

Develop general, task-agnostic mechanisms for constructing the sparse keyframe set Φ^{key} and defining the editing function f_edit that transforms global poses at selected keyframes while preserving local joint trajectories, in order to extend adaptive motion tracking from a single reference motion to a broader range of tasks.

Background

The proposed approach constructs an augmented dataset by sparsifying a single reference motion into keyframes and lightly editing selected keyframes’ global poses while keeping local joint trajectories unchanged. This keyframe-based editing is currently specified manually and in a task-specific manner, limiting scalability.

A general, automatic mechanism for selecting keyframes and defining edits would expand applicability beyond tasks with simple parametric structures (e.g., jump or strike distances), making the framework more robust and widely usable.