Learning-Enabled CPS for Edge-Cloud Computing

Luigi Capogrosso1, Shengjie Xu2, Enrico Fraccaroli1,2, Franco Fummi1, Marco Cristani1, Samarjit Chakraborty2
1 Department of Engineering for Innovation Medicine, University of Verona, Italy 2 Department of Computer Science, The University of North Carolina at Chapel Hill, USA
Paper arXiv Code
🎉 Accepted @ SIES 2024 🎉
MTL-Split teaser
Possible SC architectures that can be combined with a neural network-based controller. The raw data coming from the plant is elaborated by DNN, which extracts the features required by the control unit to compute the input feedback. Dashed red lines highlight communication between the embedded and the remote device.

Abstract

Many Cyber-Physical System (CPS), such as autonomous vehicles and robots, rely on compute intensive Machine Learning (ML) algorithms, especially for perception processing. A growing trend is to implement such ML algorithms in the cloud. However, the data transfer overhead and the delay introduced in the process necessitate some form of edge-cloud solution. Here, a part of the processing is done locally and the rest on the cloud, and how to do this partitioning is being explored in the body of work referred to as Split Computing (SC). In this position paper we explore different SC architectures and discuss their implications on controller design for CPS. In particular, we discuss the delay and state estimation accuracy of these different SC architectures and how they would impact the design of the feedback controllers using them.

BibTeX

@InProceedings{capogrosso2024learning,
    author    = {Capogrosso, Luigi and Xu, Shengjie and Fraccaroli, Enrico and Cristani, Marco and Fummi, Franco and Chakraborty, Samarjit},
    booktitle = {14th International Symposium on Industrial Embedded Systems (SIES)},
    title     = {{Learning-Enabled CPS for Edge-Cloud Computing}},
    year      = {2024},
    doi       = {10.1109/sies62473.2024.10767956},
}