CPS Design with Learning-Enabled Components: A Case Study (review)

(source: CPS Design with Learning-Enabled Components: A Case Study)
no copyright infringement intended

CPS Design with Learning-Enabled Components: A Case Study

Charles Hartsell, Nagabhushan Mahadevan, Shreyas Ramakrishna, Abhishek Dubey, Taylor Johnson, Xenofon Koutsoukos, Janos Sztipanovits, Gabor Karsai

Institute for Software Integrated Systems, Vanderbilt University

Cyber-Physical Systems (CPS [1]) integrate hardware/software components with mechanical/electronic equipment to operate in applications for robotics, avionics, smart grids, you name it. This paper presents a case-study for the design of a CPS to control the movement of an autonomous Unmanned-Underwater-Vehicle (UUV - some call it submarine drone [2, 3]).  The UUV tracks, using images from a forward-looking camera, a pipe placed on the seafloor.

Several aspects are to be considered. Firstly, the innate degree of uncertainty due to the complex interactions inside the system and between the system and the environment. The traditional control technologies show their limits and are replaced by CPSs based on Learning-Enabled-Components (LEC [4]).  Secondly, the security-critical nature of these applications: they must react correctly even to events that happen only rarely. One would say that here is a fluid space, where the only reasonable attitude is to expect the unexpected. Anyway, the design must comply to certification processes requiring safety assurance arguments backed by substantial evidence [5, 6]. Thirdly, all this complexity requires a framework supporting environment simulation and also testing in the earlier stages, including on the software models [7]. The authors used as development framework the Assurance-based-Learning-enabled CPS Toolchain (ALC [5]), a veritable engineering studio, rich in tools and workflows.

The development starts with the modeling of components and messages. ALC puts here three tools: the SysML language [8] to define the components as blocks, the Robot-Operating-System (ROS [9]) for inter-components communication, and the WebGME [10] infrastructure for instantiating the blocks, to get the whole architectonic model. Anytime when the original blocks are modified, ALC updates their instantiations automatically. For each block can be various implementation solutions. All these implementations will be evaluated to get the optimum.

Follows the LEC construction. ALC allows the developers to insert code in their blocks. For the UUV application Python was used. Data are generated using a Gazebo [11] environment simulator (the authors are currently integrating the SCENIC [12] language for data generation). ALC supports training through Artificial-Neural-Networks [4], and the authors used the supervised learning [13]. The goal is to approximate the ideal mapping function from a set of input variables to a corresponding set of output variables. The whole process is iterative.

The paper presents in a detailed manner the development cycle, and evaluates the results obtained for diverse changes of conditions (various architectures for the neural networks, various geometries of the pipeline, etc.) The concluding section identifies also possible avenues for future research, related to the formalization and quantitative evaluation of safety case arguments [14].
The authors are with the Vanderbilt University.  Paper level: industry/academia.


REFERENCES
1. Abhishek Ghosh, What is Cyber-Physical System (CPS), The Customized Windows, January 3, 2018,  https://thecustomizewindows.com/2018/01/cyber-physical-system-cps/
2. Dan Gettinger, What you need to know about underwater drones, Center for the Study of the Drone at Bard College, 2015, https://robohub.org/what-you-need-to-know-about-underwater-drones/
3. Dan Gettinger, Underwater Drones (Updated), Center for the Study of the Drone at Bard College, 2016, https://dronecenter.bard.edu/underwater-drones-updated/
4. Simon Haykin, Neural Networks and Learning Machines, Third Edition,  Pearson Education, Inc., Upper Saddle River, NJ, 2009, http://dai.fmph.uniba.sk/courses/NN/haykin.neural-networks.3ed.2009.pdf
5. Shreyas Ramakrishna, Charles Hartsell, Abhishek Dubey, Partha Pal, Gabor Karsai, A Methodology for Automating Assurance Case Generation, arXiv, 2003, https://arxiv.org/pdf/2003.05388.pdf
6. RTCA. DO-178C - Software Considerations in Airborne Systems and Equipment Certification. December 2011, https://www.qa-systems.com/index.php?id=727
7. Carlos A. Gonzalez, Mojtaba Varmazyar, Shiva Nejati, Lionel C. Briand, Yago Isasi, Enabling Model Testing of Cyber-Physical Systems, ACM/IEEE 21th International Conference on Model Driven Engineering Languages and Systems (MODELS ’18), October 14–19, 2018, Copenhagen, Denmark, https://dl.acm.org/doi/pdf/10.1145/3239372.3239409
8. OMG. OMG Systems Modeling Language (OMG SysML), Version 1.5, 2017, http://www.omgwiki.org/OMGSysML/lib/exe/fetch.php?media=sysml-roadmap:sysml_v2_requirement_support_document_2017-09-23-omg_syseng-2017-09-01.pdf
9. Morgan Quigley, Brian Gerkey, Ken Conley, Josh Faust, Tully Foote, Jeremy Leibs, Eric Berger, Rob Wheeler, Andrew Ng, ROS: an open-source Robot Operating System, ICRA Workshop on Open Source Software, 2009, https://www.semanticscholar.org/paper/ROS%3A-an-open-source-Robot-Operating-System-Quigley-Conley/d45eaee8b2e047306329e5dbfc954e6dd318ca1e
10. Miklós Maróti, Tamás Kecskés, Róbert Kereskényi, Brian Broll, Péter Völgyesi, László Jurácz, Tihamer Levendovszky, and Ákos Lédeczi. Next generation (meta) modeling: Web-and cloud-based collaborative tool infrastructure. Proceedings of MPM MPM@ MoDELS, 1237:41–60, 2014, http://ceur-ws.org/Vol-1237/paper5.pdf
11. Nathan P Koenig, Andrew Howard, Design and use paradigms for gazebo, an open-source multi-robot simulator, Proceedings of 2004 IEEE/RSJ International Conference on Intelligent Robots and Systems. Citeseer, 2004, http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.304.8999&rep=rep1&type=pdf
12. Daniel J. Fremont, Tomasso Dreossi, Shromona Ghosh, Xiangyu Yue, Alberto L. Sangiovanni-Vincentelli, Sanjit A. Seshia. SCENIC: a language for scenario specification and scene generation, Proceedings of the 40th ACM SIGPLAN Conference on Programming Language Design and Implementation, pages 63–78. ACM, 2019, https://dl.acm.org/doi/pdf/10.1145/3314221.3314633
13. Ian Goodfellow, Yoshua Bengio, Aaron Courville, Deep Learning, MIT Press, Cambridge, MA, 2016, https://www.deeplearningbook.org/
14. Rui Wang, Jérémie Guiochet, Gilles Motet, Confidence Assessment Framework for Safety Arguments,  Computer Safety, Reliability, and Security: 36th International Conference, SAFECOMP 2017, Trento, Italy, September 13-15, 2017, Proceedings (pp.55-68), https://www.researchgate.net/publication/319138629
_Confidence_Assessment_Framework_for_Safety_Arguments