Introduction

Welcome to the Innoguard MSCA Doctoral Network Project. The development of Cyber-Physical Systems (CPSs), particularly Autonomous CPSs (ACPSs) that use AI for decision-making, is becoming increasingly complex, especially as fully automated public transport is expected to grow from 30% to 70% by 2030. These systems face challenges due to their seamless connectivity, abundant computing power, and hardware heterogeneity, with software quality being crucial. Ensuring the dependability of ACPSs is critical because failures can lead to severe consequences, such as those seen in Tesla's autopilot system, linked to over 700 accidents and 17 fatalities.

Developers need to perform quality assurance on ACPSs, which includes analyzing, testing, and debugging software that interacts with hardware and assessing the quality of the AI components. This becomes more challenging with the introduction of regulations like the EU AI-Act. AI-based methods can help automate tasks like code review, testing, and debugging, leveraging deep learning models such as Large Language Models (LLMs), Generative Adversarial Networks (GANs), and deep reinforcement learning. These models have been successful in various domains and can improve software development processes.

However, deep learning models have limitations, such as bias and hallucination issues, and may require fine-tuning and hybrid methods to be effective in specific domains. Applying these techniques to ACPSs is particularly challenging due to the complex interactions between software and hardware, limited training data, and the safety-critical and security requirements of ACPSs. Therefore, a cautious approach is necessary, focusing on evaluating AI trustworthiness, uncertainty, and bias. Despite these challenges, deep learning techniques are expected to play a crucial role in developing reliable and trustworthy ACPS solutions, though their use may also lead to legal issues.

Objectives

The main goal of the InnoGuard project is to develop novel techniques, as well as methodological principles for the exploitation of AI methods such as deep learning, evolutionary algorithms, reinforcement learning, and large language models engineering, in the context of a peculiar activity of ACPS development, i.e., quality assurance. While ACPS development is a complex activity for which automation is highly desirable in various phases, including for example requirement engineering, design, or coding, our focus is on quality assurance, especially considering the high dependability requirements many of such systems have. Such ambitious objective will be possible to achieve through the following seven tangible objectives:

  • Objective 1: Realize Innovative Training Program & Development Activities specifically tailored to the ACPS context:
  • Objective 2: Automate quality assessment and evolution of ACPS behavior to ensure high ACPS trustworthiness and reliability
  • Objective 3: Enhancing Dependability: Real-time Security, Privacy, and Uncertainty Handling in ACPS Operation
  • Objective 4: Enhancing AI trustworthiness and quality improvement approaches through novel quality engineering methods
  • Objective 5: Enhance the environmental sustainability of both ACPSs and the LLMs supporting their engineering
  • Objective 6: Validate the relevance and cost-effectiveness of the project services in open source contexts
  • Objective 7: Effectively communicate and disseminate project results

Partners

Our project is a collaboration between several prestigious institutions, including:

  • Mondragon University (Spain)
  • Universita degli Studi del Sannio (Italy)
  • Simula Research Laboratory (Norway)
  • Oslo Metropilitan University (Norway)
  • Vrije Universiteit Amsterdam (Netherlands)
  • University of Malaga (Spain)
  • Zurcher Hochschule Fur Angewandte Wissenschaften (Switzerland)

Work Packages

Our project is divided into several work packages:

  • Work Package 1: Project Management, Governance and Coordination
  • Work Package 2: Network wide and Individual Training
  • Work Package 3: Research in Design Time Quality Engineering of ACPS
  • Work Package 4: Methods for Monitoring, Self-healing, and Self-adaptability of ACPS in Diverse Field Context
  • Work Package 5: Trustworthiness of Quality Engineering Methods
  • Work Package 6: Dissemination and Exploitation

Timeline

The project is scheduled to run from 1st of September 2024 to August 2028. However, it is expected that doctoral candidates will join their corresponding institutions by February 2025. Each project is expected to last 3 full years.

Contact Information

If you have any questions or would like to get involved, please contact us at:

Email: aarrieta@mondragon.edu