Reliable and Sufficient Guidance Construction for Professional Motion Editing

Construct reliable and sufficient multimodal guidance for professional motion editing, including the acquisition and verification of external references that accurately specify the target motion technique and temporal structure.

Background

Professional motion editing requires technically precise motion descriptions, including body coordination, support relationships, temporal ordering, and key poses. Existing video-editing task formulations may accept text instructions and reference videos, but they generally assume that a suitable reference is already available rather than addressing how to acquire and verify one.

External search can provide candidate demonstrations, but relevance alone does not ensure that a reference correctly depicts the requested professional motion or contains a clear, complete, and usable segment. The paper therefore identifies the construction of reliable and sufficient guidance—including both guidance acquisition and verification—as an unresolved problem that motivates the MotionCoach framework.