AnyMo: Geometry-Aware Setup-Agnostic Modeling of Human Motion in the Wild

Baiyu Chen, Zechen Li, Wilson Wongso, Lihuan Li, Xiachong Lin, Hao Xue, Benjamin Tag, Flora Salim

Summary

Treats wearable setup variation as structured body-surface geometry: dense physics-grounded IMU simulation on Nymeria mesh → ST-GCN masked cross-view predictive contrastive pretrain → PQ full-body IMU tokens aligned to Qwen2.5-0.5B for open-vocabulary HAR, retrieval, and captioning. Strong zero-shot gains across 14 unseen HAR sets vs UniMTS / IMU2CLIP / IMUGPT / NormWear / large LLM prompting.