Skip to content

Toward Dynamic Optimization

Combining AI and Evidence-based Health Design for the Elderly

Therapeutic gardens can provide potential health benefits for the elderly. The Evidence-based Health Design in Landscape Architecture (EBHDL) process model can help deliver stated health outcomes. With the development of artificial intelligence, the growth of computer-aided design can improve the EBHDL process to fill the gaps.

This research project aims to explore how artificial intelligence combined with the EBHDL process model can contribute to the therapeutic garden design for the elderly. The proposed framework was the first attempt to apply different GANs and to determine their applicable characteristics to the therapeutic garden design, making evidence-based health design for the elderly towards dynamic and scientific optimization.

This project has won the ASLA 2023 Student Award.


Trained GANS models that can generate garden plans,

Conducted experiments related to therapeutic gardens

Graphize the results and submit them to the competition


Therapeutic gardens provide health benefits for the elderly. The EBHDL model helps deliver stated health outcomes but is time-consuming and costly.
With the development of AI, the growth of computer-aided design such as GANs, is possible to fill the gaps in the EBHDL process.
The framework combines the EBHDL process model with AI algorithms including 4 steps: 1) Evidence collection; 2) Programming; 3) Design; 4) Evaluation.
We collected related health challenges, knowing how the elderly can benefit from nature exposure and how can be used to support their health.
We collect site-related evidence and explore how the design can support the intervention in promoting the elderly’s health within the selected site.
425 samples of therapeutic gardens are collected, classified, pre-processed, and translated into computer ‘language’ to build 2 datasets for training.
We trained four kinds of GANs using paired and unpaired datasets to generate garden layouts respectively. Finally, twelve models are trained.
12 garden layouts generated by the well-trained models are evaluated by a series of criteria. The highest score layout is selected for further design.
Designers systemize evidence for improving basic elements, promoting walkability, and adding therapeutic elements based on design aims and criteria.
The therapeutic garden is conceptualized and designed based on the selected garden layout and the developed design strategies.
Approoriate plant combinations, well-designed blue space, rest space, and plaza can alleviate the elderly’s fatigue and anxiety about their health.
Walkability-related elements like barrier-free facilities, walkways, and shelters are provided for the elderly to engage in outdoor activities.
A garden with therapeutic elements like ornamental plants, rest facilities, etc., could offer communication and sensory stimuli opportunities.
We used wearable devices and questionnaires to evaluate health benefits in 3D virtual gardens. The new knowledge help adjust the further design.
This research develops an interdisciplinary, systematic, and dynamic work process between humans and computers, contributing to health design.