Evolutionary Exploration of Mechanical Assemblies in VR

Abstract

Due to the maker movement and 3D printers, people nowadays can directly fabricate mechanical devices that meet their own objectives. However, it is not intuitive to identify the relationship between specific mechanical movements and mechanical structures that facilitate such movements. This paper presents an interactive system that can enable users to easily create and experiment with desired mechanical assemblies via direct manipulation interfaces in virtual reality, as well as to intuitively explore design space through repeated application of the crossover operation, which is used at the core of the genetic algorithm. Specifically, a mechanical assembly in our system is genetically encoded as a undirected graph structure in which each node corresponds to a mechanical part and each edge represents the connection between parts. As the user selects two different mechanical assemblies and commands the crossover operation, each of their corresponding graphs is split into two subgraphs and those subgraphs are recombined to generate the next-generation mechanical assemblies. The user can visually examine new mechanical assemblies, save assemblies that are closer to objectives, and remove the others. Based on our experiments, in which non-expert participants were asked to achieve a challenging design objective, it was verified that the proposed interface exhibited significantly effective performance.

Preface to “Swarm and Evolutionary Computation—Bridging Theory and Practice”

“…Kim and Lee [10] suggest an interactive GA system that can allow users to easily create and experiment with desired mechanical assemblies, which are encoded as undirected graphs, via direct manipulation interfaces in virtual reality, and to intuitively explore design space by repeatedly applying the proposed crossover operator.…”