Imagine living in a cool, green city flush with parks and threaded with footpaths, bike lanes, and buses, which ferry people to shops, schools, and service centers in a matter of minutes.
That breezy dream is the epitome of urban planning, encapsulated in the idea of the 15-minute city, where all basic needs and services are within a quarter of an hour’s reach, improving public health and lowering vehicle emissions.
Artificial intelligence could help urban planners realize that vision faster, with a new study from researchers at Tsinghua University in China demonstrating how machine learning can generate more efficient spatial layouts than humans can, and in a fraction of the time.
Automation scientist Yu Zheng and colleagues wanted to find new solutions to improve our cities, which are fast becoming congested and concrete.
They developed an AI system to tackle the most tedious, computational tasks of urban planning – and found it produces urban plans that outperform human designs by about 50 percent on three metrics: access to services and green spaces, and traffic levels.
Starting small, Zheng and colleagues tasked their model with designing urban areas only a few square kilometers in size (about 3×3 blocks).
After two days of training, and using several neural networks, the AI system searched for the ideal road layouts and land use, to fit with the concept of the 15-minute city and local planning policies and needs.
While Zheng and colleagues’ AI model has some features to scale up its use to planning larger urban areas, designing an entire city would be infinitely more complex. Drafting a neighborhood consisting of 4×4 blocks contains twice as many planning decisions as 3×3 blocks, the researchers estimate.
But automating even a few steps in the planning process could save huge amounts of time: the AI model computed in seconds certain tasks that took human planners between 50 to 100 minutes to work through.
Automating the most time-consuming tasks of urban planning would free up planners to focus on more challenging or human-centric tasks, such as public engagement and aesthetics, the researchers say.
Rather than AI replacing people, Zheng and colleagues envisage their AI system working as an ‘assistant’ to urban planners, who could generate concept designs that are optimized by the algorithms, and reviewed, adjusted, and evaluated by human experts based on community feedback.
This last step is central to good design, Massachusetts Institute of Technology (MIT) research scientist Paolo Santi writes in a commentary on the study.
Urban planning is “not merely an allocation of space to buildings, parks and functions, but the design of a place where urban communities will live, work, interact and, hopefully, thrive for a very long time,” he writes.
Comparing their human-AI workflow to human-only designs, Zheng and colleagues found the collaborative process could increase access to basic services and parks by 12 and 5 percent, respectively.
The researchers also surveyed 100 urban designers, who were unaware of whether the plans they were asked to choose between were generated by human planners or AI. The AI won substantially more votes for some of its spatial designs, but for other plans, there was no clear preference among survey participants.
The true test would of course be in the communities built from those plans, measured by the reduction in noise, heat, and pollution, and improvements in public health that better urban planning promises to bring.
The study has been published in Nature Computational Science.
