Many automated space layout (ASL) studies attempt to take advantage of the machine to complete time-consuming assembly building layouts by taking spatial adjacency as an optimization objective. Although the layout generated by machines can satisfy adjacency needs and topological relationships, the architect still needs further a lot of manual modifications to obtain a customized layout. Many ASL methods focus on optimizing fixed optimization objectives, they hardly use the architect’s decisions as dynamic optimization parameters to constantly approximate the architect’s preferences. Motivated by this, we provide an architect-machine collaborative design (AMCD) workflow that combines the strength of the architect’s visual recognition and reasoning ability with the strengths of machines to enhance the ASL process. AMCD optimizes the adjacency relationship in real-time and simultaneously takes the architect’s design decision as dynamic optimization parameters to jointly calculate the design layout that satisfies the adjacency relationship and meets the architect’s preference. Architects can dominate the layout results with experience and visualization of adjacency performance by manually modeling and modifying machine errors during iterations. A case study demonstrates that our workflow can initially complete different customized design tasks. The AMCD workflow demonstrates the potential to enhance architects’ decisions and optimize layouts in contrast to fully architect-designed and fully automated machine-designed layouts. In the future, our framework has the potential to provide labeled datasets to train machine learning models for completing complex building design tasks forming an ArchitectMachine Collaborative tool.

1

Ximing Zhong

 

ximing.zhong@aalto.fi

Shengyu Liu

Fujia Yu

fujiashangcheng@gmail.com

Yunhao Zhong

Beichen Xu

Conference Video