Proceedings of the

The 33rd European Safety and Reliability Conference (ESREL 2023)
3 – 8 September 2023, Southampton, UK

Adversarial Multi-Agent Reinforcement Learning for Fault-Tolerant Design of Complex Systems

Joachim Grimstada and Andrey Morozovb

Institute of Automation and software systems, University of Stuttgart, Germany.


With the increasing complexity of systems, the design and operation of these systems have become increasingly challenging, necessitating greater expertise from engineers and operators. Ensuring the safety and reliability of such systems is crucial, however, traditional design methodologies may not prove adequate to manage the growing complexity of systems. This paper proposes a conceptual approach to designing fault-tolerant complex systems. The approach extends Model-Based System Engineering (MBSE) with zero-sum game models. These models allow Adversarial Multi-Agent Reinforcement Learning (Adv-MARL) techniques to explore various strategies, and assess outcomes. They also have the potential to identify vulnerabilities that can be addressed by iteratively refining the system design.

Keywords: Model-based system engineering, Adversarial multi-agent reinforcement learning, Fault-tolerant design, Zero-sum game.

Download PDF