Search papers, labs, and topics across Lattice.
MIT CSAIL
Tsinghua AIBeijing Normal鈥揌ong Kong Baptistgithub.comHKUSTJD Logisticspkucs-Ltf/LongAV-CompassRutgersMay 21, 2026arXiv:2605.22086The paper introduces GenHAR, a framework for cross-domain Human Activity Recognition (HAR) that learns domain-invariant sensor representations to improve generalization. GenHAR tokenizes sensor data and learns correlations among frequency sensor channel dimensions, while also incorporating selective masking and efficient attention to improve efficiency. Experiments on real-world datasets show that GenHAR outperforms state-of-the-art methods by 9.97% in accuracy and reduces Floating Point Operations by 6.4 times, with successful real-world deployment at a logistics company.
Achieve 9.97% higher accuracy in cross-domain human activity recognition while simultaneously reducing computation by 6.4x with a new sensor data tokenization and attention mechanism.
Human Activity Recognition (HAR) has shown remarkable effectiveness in various applications, such as smart healthcare and intelligent manufacturing. However, a major challenge faced by HAR is the distribution shift across different sensor data domains, which often leads to decreased performance when deployed for real-world applications. To address this issue, this paper introduces GenHAR, a novel framework designed to mitigate the domain gap by learning domain-invariant sensor representations. GenHAR aims to enhance the generalization capabilities of HAR on target domains purely with data from the source domain. The key novelty of GenHAR lies in two aspects. Firstly, GenHAR tokenizes sensor data and learns correlations among frequency sensor channel dimensions to improve the robustness of HAR models. Secondly, GenHAR improves the efficiency via selective masking and an efficient attention mechanism. We conduct a systematic analysis of GenHAR by comparing it with state-of-the-art HAR methods on real-world human activity datasets. Results show that GenHAR outperforms state-of-the-art methods by 9.97% in accuracy, and reduces Floating Point Operations by 6.4 times. Moreover, we deploy GenHAR at a leading logistics company in 4 cities, and have detected 2.15 billion real-time activities. We release our code at: https://github.com/Sensor-FoundationModel/GenHAR.
