Motion Capture is Not the Target Domain: Scaling Synthetic Data for Learning Motion Representations
Firas Darwish, George Nicholson, Aiden Doherty, Hang Yuan · 2026
Read Motion Capture is Not the Target Domain: Scaling Synthetic Data for Learning Motion Representations by Firas Darwish, George Nicholson, Aiden Doherty, Hang Yuan (2026) 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-2602-11064.txt.
Description
Real motion data is scarce, yet wearables need robust activity models. Firas Darwish and three co-authors test built motion data for pretraining, built from motion-capture-derived forms. They train time-series models on built motion and check transfer to varied human activity tasks.
Built pretraining helps when mixed with real data or scaled up enough. By contrast, large-scale motion-capture pretraining alone adds little, since its signals mismatch wearable data. The paper clarifies sim-to-real limits for motion: built data is a useful lever, but only when its domain gap to deployment sensors is faced head-on.
Catalog record
| CREATOR | Firas Darwish, George Nicholson, Aiden Doherty, Hang Yuan |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-11064