DCAF: Cross-Attention Guided Dynamic Feature Fusion from Robotic Anomaly Detection to Position Accuracy Modeling

Published

August 17, 2025

Author(s)

Guixiu Qiao, Pavlo Piliptchak, Hui Liu, Ran Jin, James Moore, Daniela Sawyer, Yingyan Zeng

Abstract

In industrial applications, inconsistent sensor configurations lead to significant differences in data types collected from different devices. Additionally, the lack of historical data in new scenarios complicates model development and limits the applicability of existing methods. Current transfer learning approaches rely heavily on static feature extractors, which fail to dynamically adjust to specific relationships between modalities or samples. These methods struggle to capture inter-modal associations effectively, resulting in insufficient information utilization. They may also extract features irrelevant to the target task, harming the training process and reducing predictive performance. To address these challenges, this paper proposes a Dynamic Cross-Attention Feature Fusion (DCAF) approach. By calculating attention weights tailored to each target domain sample, DCAF extracts the most relevant source domain features and generates dynamic fused representations. The proposed approach enables sample-specific feature selection and fine-grained domain alignment, effectively overcoming the limitations of static feature alignment in traditional transfer learning. It is particularly suited for small-sample and heterogeneous data scenarios. Experimental results on the Sheffield and NIST datasets demonstrate the effectiveness of DCAF, showcasing strong adaptability and robustness, providing an efficient solution for domain adaptation and multimodal fusion.

Conference Dates

August 17-21, 2025

Conference Location

Los Angeles, CA, US

Conference Title

IEEE International Conference on Automation Science and Engineering (CASE)

Pub Type

Conferences

Download Paper

Local Download

Keywords

robotics, Anomaly detection, dataset, machine learning

Citation

Qiao, G. , Piliptchak, P. , Liu, H. , Jin, R. , Moore, J. , Sawyer, D. and Zeng, Y. (2025), DCAF: Cross-Attention Guided Dynamic Feature Fusion from Robotic Anomaly Detection to Position Accuracy Modeling, IEEE International Conference on Automation Science and Engineering (CASE), Los Angeles, CA, US, [online], https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=959674 (Accessed May 20, 2026)

Additional citation formats