Researchers have introduced TRACE, a framework for conversational editing of humanoid motion. The work, posted on arXiv, targets language-conditioned motion generation, where users describe movements in natural language. The central problem, the abstract states, is connecting an underspecified request to concrete motion structure.

TRACE stands for Typed Reference-Aware Conversational Editing. It builds on motion diffusion models, which generate motion from text. The 'typed reference-aware' component appears designed to let users refer to specific parts or aspects of a motion during a back-and-forth conversation, rather than issuing a single one-shot command.

Because the abstract is brief, the paper's full details—such as evaluation results or comparisons to other methods—are not available from this source. The contribution, as described, is a step toward more interactive and precise control over generated humanoid movement.