A novel deep generative model based on imaginal thinking for automating design

Abstract

Automating design faces a thorny problem: insight modeling based on knowledge and experience. In particular, it is difficult for artificial intelligence to perform incomplete conditional reasoning. The deep generative model (DGM) is an emerging approach of machine learning, which typically uses deep networks to learn from various data sets and synthesize new designs. This paper proposes a novel DGM based on imaginal thinking to realize the creative leap from the invisible functional domain to the concrete physical domain. An experiment is conducted to verify the effectiveness of the proposed model in designing wheels for mobile robots in granular media.