Evolvable Automata for Engineering Design Exploration paper – (September 2026)

This paper was presented at the UK Computational Intelligence Workshop 9-11th September 2026 by Simon Hickinbotham and supported by Professor Andy Tyrrell

Abstract. Automation of the process of engineering design offers the potential to explore the range of design solutions more thoroughly than is possible manually, making new discoveries and delivering efficient solutions to human endeavour. However, the problem of representing real physical objects in design systems remains challenging. Evolutionary approaches offer many advantages due to their ability to search with in complete knowledge, but they generally require that three-dimensional space be represented on a one-dimensional genome, which can significantly affect the potential diversity of solutions. This study harnesses recent advances in continuous and particle automata for engineering design exploration. An engineering design problem is presented to the system, which uses a representation of the problem as a scalar field in the continuous automaton to deliver improved design solutions. A simple but novel evolutionary algorithm, designed to maintain diverse solutions is used to drive this improvement. Previous approaches to evolving engineering designs used either discrete ball-and-stick representations or continuous fields. The method presented here brings the advantages of both techniques together for the first time. Thus a better representation of engineering design problems is developed, which has a wide range of applications in industry.