Foundation models aim to deliver broad generalization, but automated manufacturing differs greatly from the internet-scale data that general-purpose LLMs/VLMs are trained on. Manufacturing data is often sparse, small-sample, and highly heterogeneous (e.g., 1D sensor signals, 2D images, and 3D CAD models), so directly applying off-the-shelf models is frequently insufficient. This special issue focuses on developing and rigorously adapting foundation models under manufacturing constraints, including building robust models despite limited failure data and complex physical dynamics.

This special collection has two objectives: (1) to assess current capabilities and limitations of foundation models in automated manufacturing, and (2) to advance model design and adaptation to better meet real-world manufacturing challenges. Submissions are encouraged to provide critical insights into limitations observed in practice and to propose new architectures, training paradigms, or system designs that address these gaps.

List of Topics

• Agentic AI and Embodied Reasoning
• From Semantic to Metric
• World Models for Manufacturing
• Cross-embodiment learning
• Industrial-Grade Trustworthiness
• Physics-informed foundation models

• Methodologies for developing foundation models using small, sparse, or unbalanced manufacturing datasets

Organizers

• Binglu Wang, Dr., Northwestern Polytechnical University, CN, E-mail: wbl@nwpu.edu.cn
• Xiaoyu Guo, Dr., City University of Hong Kong, HK, E-mail: xiaoyguo@cityu.edu.hk
• Hongsheng He, Dr., The University of Alabama, USA, E-mail: hhe11@ua.edu
• Zuowei Zhang, Dr., Northwestern Polytechnical University, CN, E-mail: zhangzuowei@nwpu.edu.cn
• Chenguang Yang, Dr., University of Liverpool, UK, E-mail: Chenguang.Yang@liverpool.ac.uk
• Huaping Wang, Dr., Beijing Institute of Technology, CN, E-mail: wanghuaping@bit.edu.cn
• Jun Yang, Dr., Loughborough University, UK, E-mail: j.yang3@lboro.ac.uk
• Juan José Rodriquez-Andina, Dr., University of Vigo, Spain, E-mail: jjrdguez@uvigo.es