AI Mariupol

Nehal Jain and John Li, A. Alfred Taubman College of Architecture + Urban Planning

The increasing developments in Artificial Intelligence (AI) and its applications in various fields have revolutionized the way things are done. The architecture and urban design fields are no exceptions, and this thesis aims to explore the potential of AI in streamlining the city planning and design process and providing improved outcomes.

For our thesis, we chose Mariupol, Ukraine as a site due to the recent conversations and the dire need of the city to be reconfigured, designed and planned. The call of action is urgent and artificial intelligence can be used as a tool to aid in rebuilding Ukraine. Although this project is highly political and the war is ongoing, we believe it is valuable because it is important to have public discussions about how the architectural community can participate in the efforts to rebuild an independent Ukraine. The recovery of public infrastructure will be necessary as one of the first steps towards a “new normal”.

Our thesis includes collaboration within four disciplines: Architecture, Urban Design, Artificial Intelligence(Computer Science) and Socio-cultural studies. Even though we chose Ukraine as a case study site for our project, it is important to consider that the tools that we are aiming to build throughout the course of our thesis are scalable and can be used in a different context as well, whether it’s about designing a building, one block of the city or entire city.

STEP 1: We will use annotated satellite images taken from other global cities for our dataset. These images will be annotated with information about infrastructure, walkability, transportation, and other features that can contribute to a city’s quality of living and success. These images will be around 4 miles X 4 miles in terms of area of the city that will be focused. However, this scale can be modified and be applied to a larger city scale or a smaller building scale. One thing we want to acknowledge is that this war is rooted in imperialism, and we are not blind to the biases and nature of media and artificial intelligence. Through our dataset and approach, we hope to both acknowledge and critique this fact. After gathering this dataset, we plan to feed the data to a neural network, in our case StyleGan2 and Runway ML. This process will then generate diffusion city plans and after choosing a plan that is well suited to the context of Mariupol, we will be using another neural network called Styletransfer to apply the plan onto the specific geographical base of Mariupol. We will develop and refine the city plan then to include cultural factors that may be difficult to address using AI in earlier steps. Since the images generated from these platforms are relatively low in resolution, we will be using to increase the resolution of the generated city.

STEP 2: We will then be zooming onto the center of Mariupol (where the Donetsk Academic Regional Drama Theatre was), and design the area around it and the city plaza, The city plaza will be focused on the culture and the demographics of Mariupol, and the design of the plaza and the areas around will tend to focus on the social cohesion that would be important for the city in the future. We will be using design strategies as well as AI tools (Style Gan and Runway ML)to design this plaza and the surrounding areas.

At the end of this project, we aim to produce (1) a large printed 4 mile x 4 mile infrastructure plan of Mariupol(around 4ft X 6ft in print), (2) a drawing that shows the city center and the plaza of Mariupol (around 5ft X 5ftin print), and a (3) large scale physical model (around 5ft X 5 ft X 3 ft) that shows the scope of our work as well. We also plan to print our thesis and the entire project information (that includes the process and supplemental information) in the form of a book. We plan to print around 12 books that can be distributed to the reviewers that come to Taubman at the final review.

We plan to finish our thesis and present it to reviewers on April 21/22 during our final review. This project will also be presented on April 28 at the Taubman Architecture Liberty Annex. We also plan to showcase this project in the potential conference in next semester (Fall 2023) about Rebuilding Ukraine at Taubman College of Architecture and Urban Design. This project will also be published on the website for our studio course, as well as students’ individual websites so the architectural community has access to these resources.

An obvious benefit of this project is that it acts as a prototype tool that can be used to not only rebuild other war areas, but also areas that lack infrastructure. Because the AI is not generating the diffusion city on an open field but to the specific site, it can be applied to any city. Further application includes the ability to adjust different factors such as walkability and filtering by specific infrastructures as well as the ability to adjust the scales of the project depending on the context. In the StyleGan component, few loss functions can also be added to make the plan more specific to the context. We plan to add one such loss function (making sure the parks are within a walking distance) to our StyleGan code. In order to use StyleGan and Style Transfer, we would use Google Colab Pro (a membership-based service). We, as architecture students, are interested in the discipline and the intersection of AI and design. Through this project, we are trying to gain a deeper understanding on the role of Artificial Intelligence in our lives, and especially using it as a design tool for the social benefit. We plan that this tool will lead to a healthy collaboration between computer science and architectural discourses and will lead to a deeper analysis of intersection between AI, architecture, urban design, and social studies which is often not very visible in the field.

Project Update

Nehal and John completed their thesis Chaos to Cohesion: The Role of Artificial Intelligence in Urban Planning and Design.

The project definitely challenged us to broaden our perspective on architectural methods while including artificial intelligence, history and sociology as important factors. Designing for a war-torn city as it already is challenging and can result into multiple debates, having a broader perspective helped us throughout the process. AI and architectural design were more intertwined than we had expected. We were able to reach out to Alexandra Carlson, Research Scientist as well as attended various workshops, one of the most important being lectures by Harvard Ukrainian Research Institute where there were panelists from various fields including but not limited to architecture, politics and history.

The field of AI has been emerging recently and has opened carious employment opportunities. Knowing how to navigate in this field and using the resources that can help with the automatization of architecture has helped open paths to various firms and jobs that are approaching towards this area and helps us be on top of the changes that can reform architecture in the near future.

We would like to express our sincere gratitude to Dr. Matias del Campo for his unwavering support and guidance throughout the entirety of our project. As one of the best thesis advisors and mentors, he provided invaluable feedback, insight, and motivation that were essential to the success of our work. Without his expert guidance, this project would not have been possible.

We would also like to extend our thanks to Dr. Oksana Chabanyuk for her constant guidance throughout our project and her unique and first-hand perspective on architecture in Soviet Ukraine. Her knowledge and expertise added depth and nuance to our work, and we are grateful for her input.

We must acknowledge Devishi Suresh, MS Student in Electrical and Computer Engineering, for her technical expertise in machine learning. Her contributions were crucial in bringing our project to fruition.

Additionally, we would like to thank the ArtsEngine and AiiR Grant Committee for their support, which helped us to fund our thesis project.