HALO: A Heterogeneity-Aware Language-Aligned IMU Foundation Model for Open-Set Human Activity Recognition

Zihan Ding, Liyu Zhang, Xiaomin Ouyang (HKUST)

Summary

IMU FM handling rate/channel/placement heterogeneity + open-vocabulary HAR via synonym-aware soft contrastive alignment to Sentence-BERT. ~35M trainable params; +13.7 pp zero-shot open-set Acc vs MOMENT on 5 held-out sets; severe OOD still collapses.