A Glimpse on STAR’s Reference Model for Trusted AI Systems
The H2020 STAR project is developing technologies for trusted Artificial Intelligence (AI) systems in production lines. Specifically, the project develops technologies that ensure:
- Secure and Reliability for Industrial Data: STAR ensures that AI systems operate over reliable industrial data based on technological solutions (e.g., data provenance, tampered-proof data management) that alleviate the inherent unreliability of industrial data.
- Secure and Trusted AI algorithms: STAR develops AI technologies that secure the operation of the AI systems and algorithms that they comprise. In this direction, the project implements cyber-defence strategies that protect and defend AI systems from malicious security attacks. STAR focuses primarily on defences against cyber-security attacks. Physical security attacks are applicable to some STAR systems (e.g., robotics systems used in the project), yet they are not considered in the scope of the project.
- Trusted Human AI interactions: STAR focuses on the implementation of trusted interactions between humans and AI systems. On the one hand, the project ensures that AI systems are transparent and explainable to humans towards boosting their acceptance and adoption. On the other, the project focuses also on safe and trusted interactions between humans and AI systems in scenarios like human robot collaboration.
- Safe AI systems: STAR includes research towards ensuring the safety of autonomous AI systems such as mobile robots. It focuses for example on the secure placement and movement of Autonomous Mobile Robots (AMRs) in the context of the plant. These systems fall in the broader scope of the safe operation of autonomous systems.
The project has recently produced an initial version of the architecture of its platform, which illustrates the structuring principles and building blocks of a STAR-compliant trusted AI system, which exhibits the above-listed properties. The STAR architecture follows principles of standards-based reference architectures for industrial systems such as the Industrial Internet Reference Architecture (IIRA) of the Industrial Internet Consortium and the Industrial Internet Security Framework (IISF). In-line with the IISF, the STAR architecture specifies its security and safety functionalities as a set of cross-cutting functions that are applied to digital manufacturing platforms towards boosting their security and trusted end-to-end i.e., from the devices and the communication end points to the industrial applications. Likewise, in-line with the IIRA, the main functionalities of the STAR platform can be clustered in three main categories or domains according to the IIRA terminology. These three domains are illustrated in the following figure, which provides a high-level reference model for the functionalities of the STAR platform.
Specifically, the three domains are as follows:
- Cybersecurity Domain: Comprises functionalities that are destined to ensure the reliability and security of industrial data, as well as of AI algorithms that are trained and operational based on them. The functionalities of these domains support and reinforce the trustworthiness of the project’s functions in the other two domains. This domain includes STAR’s AI cyber-defence strategies, STAR’s blockchain based data provenance techniques, as well as the project’s security risk assessment and security policy management results.
- (Trusted) Human Robot Collaboration Domain: Provides functionalities for the trusted collaboration between human and robots. Leverages cybersecurity functionalities, while being used to reinforce functionalities in the safety domain as well. STAR’s simulated reality, active learning and human digital twin systems fall in this domain.
- Safety Domain: Ensures the safety of industrial operations, including operations that involve workers and/or automation systems. For instance, functionalities in this domain reinforce worker safety, while catering for the safe operation of AMRs in industrial sites. The STAR’s results about the safe placement of mobile robots in industrial plants, the workers’ fatigue monitoring systems, as well as various reinforcement learning techniques fall in this domain.
As illustrated in the figure the functionalities of all domains depend on AI algorithms, including Explainable Artificial Intelligence (XAI) techniques. As such they depend on the STAR AI platform and on the XAI models developed on top of it. XAI plays an instrumental role for the operation of the STAR platform, as it supports defence strategies (in the cybersecurity domain), data generation for simulated reality and active learning functionalities (in the human robot collaboration domain), as well as the development of human digital twins (in the safety domain).
The detailed specification of the STAR reference architecture specifies the building blocks of each one of the above techniques, along with the interactions and interfaces between them. As such it boosts the development of trusted AI systems in general, including STAR compliant systems. For more information about the STAR architecture, you can request our respective deliverable. For more information on the individual technologies and building blocks of the STAR platform, please refer to other blog posts on the STAR web site or to the project’s Open Access Book.
To remain up-to-date about the project’s outcomes, please follow us:
By: Dr John Soldatos, INTRASOFT International