Salsa as a Nonverbal Embodied Language– The CoMPAS3D Dataset and Benchmarks

Bermet Burkanova, Payam Jome Yazdian · 2025

Read Salsa as a Nonverbal Embodied Language– The CoMPAS3D Dataset and Benchmarks by Bermet Burkanova, Payam Jome Yazdian (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-19684.txt.

Description

Bermet Burkanova and co-authors build a dataset and test suite for a subtle robot skill: dancing with a human partner. Social robots have to read a partner's movement, intent, and skill level through the body alone. Existing benchmarks score motion with physics-style measures that cannot tell whether a generated follower's steps speak the shared language of a dance or suit the partner's level.

Their resource, CoMPAS3D, uses improvised partner salsa, a form with a known move vocabulary and judging criteria. It holds three hours of motion capture from eighteen dancers, from beginners to professionals, with over 2,800 expert-labeled segments noting moves, errors, and style. Three benchmarks cover move recognition, skill estimation, and generating the follower's part. Fine-tuned vision-language models scored well on true motion, while two generation systems failed in ways the new metrics expose and human judges confirm.

Catalog record

CREATORBermet Burkanova, Payam Jome Yazdian
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE55 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-19684

Browse the full library · About the Tower of Babel