Outlooks and Challenges in Urban Design X Data Science Machine in the Urban Design Loop

From Design Computation
(Redirected from DCIO.2021.DDJD8821)
Jump to: navigation, search
DCIO2021-Logo.png
DC I/O 2021 Paper by STEPHEN LAW.


DCIO2021 S-Law.png


Abstract

We are living in an age where data and data science methods are pervasive. The question is how do we adapt modern data science and machine learning methods into the urban design process (Carmona 2009; Hartz-Karp et al 2020). Urban design is the design of towns and cities, streets and spaces and its process is interdisciplinary and iterative; first in analysing its context, second in understanding the site, third in developing designs and fourth in evaluating the design collectively and repeating this process until the next stage in the development process. In this opinion piece, we will illustrates examples on how data science methods can be adapted into each stage of the urban design processes thereby giving the generic notion of “machine-in-the-urban-design-loop” or “machine-assisted-urban-design”; i. in discovering knowledge from a vast array of contextual data; ii. digital reconstruction for site analysis, iii. generative methods in design development and, iv. forecasting methods for design evaluation (as illustrated in fig1). We will conclude by describing some potential opportunities and challenges for this integration.


Knowledge Discovery, Design development and Evaluation

Knowledge Discovery from a vast array of contextual data

Knowledge discovery or data mining refers to the process of generating new insights from databases (Fayad et al 1996). In the context of urban design, modern data science methods allows for the translation of both structured data such as tabular data (eg. census) and unstructured data such as imagery (eg. Remotely sense satellite imagery), point cloud (eg. LiDAR), and sequential data (eg. textual data) into new insights and knowledge from the environment. Early examples include applying unsupervised methods for geodemographic classification (Harris et al 2005; Singleton and Longley 2015), the use of space syntax, and geographical methods for calculating accessibility (Hansen 1959; Hillier and Iida 2005) and the calculation of natural vegetation index from multi-spectral remotely sense satellite imagery (Tucker 1979; Pettorelli et al 2005). Being able to overlay different layers of contextual information can help designers make and justify certain design decisions. More recent examples on retrieving contextual information using machine learning methods; include the recognition of active urban frontage from street images (Law et al 2020), the tracking of pedestrians and vehicles from 3D point cloud data (Rangesh et al 2019) and the mining of opinion for different neighbourhoods in New York City using natural language processing (Hu et al 2019). The former two can identify problems (the lack of active frontages) and the pedestrian condition (the number of pedestrians) of the built environment. While the latter can help understand the perception of the environment (sentiment of the environment). This new type of information extracted from novel data sources can complement existing data layers into providing additional contextual insights for urban design practices which can be integrated into urban and building information systems.

Digital reconstruction for site analysis and augmented design

The second topic is the use of advance digital reconstruction methods for site analysis and augmented design. This opportunity comes from two sources. The first is the increasing availability of omni-directional 2D image and 3D point cloud sensors on mobile devices (Guan et al 2015) which in the past have been constraint for professional uses due to both the costs and expertise on collecting/handling/ processing the data. The second is advancements in neural rendering where highly accurate 3D environments can be re-constructed from a sparse set of 2D images from different angles. (Mildenhall et al 2020) Both the ease in mobile reality capture and quantitative reconstruction advancements will allow urban designer to quickly capture the reality of the site. By coupling with augmented reality (e.g. Oculus ) and augmented design tools (eg. Dreams on PS5 ), these technologies can allow designer to better situate themselves for immersive site analysis and 3D digital urban design. The training of future urban designer to use these technology would be an important facet for the future of design education.


Design Development using generative methods

The third topic is the application of generative methods in urban design development. Generative computer-aided urban design is a well-studied topic that can be categorised into rule based methods and optimisation methods. (Koenig et al 2019; Wootman et al. 2017). There is a growing interests on using machine learning techniques such as Generative Adversarial Networks (GANs) in design computation. These generative models which can produce highly realistic images (Goodfellow et al 2014; Karras et al 2020) consists of a generator that synthesize images and a discriminator that improves the image quality iteratively. Examples include its use to generate street network patches such as StreetGAN (Hartmann et al. 2017), and the generation of floor plans in architecture such as ArchiGAN (Chaillou 2020). However, the problem with image synthesis methods such as GAN is that these methods differ to how a designer design and thus the synthesize image can often lack realism. As a result, there had been a couple of research direction that could potentially help in the generative design process. One is stroke-based neural painting methods which produces strokes when rendering. Examples include applying recurrent neural network to produce vector sketches or applying reinforcement learning (RL) agent that learns to paint (Huang et al 2019; Tian et al 2020). The latter methods have found great success in problems such as game playing and medical discovery (Silver et al 2016; Jumper et al 2021). Adapting such techniques in design, Han et al (2022) have recently trained a RL agent to create an array of 3D volumetric masterplanning scenarios (maximising certain design criteria) in assisting the design process.

Design Evaluation using predictive forecast models

The fourth and final topic is to use supervise learning models to learn a function that maps an input to an output performance metric, for design evaluation. Such models can bring objectivity and evidence into the evaluative design process rather than by heuristics. Examples include the use of imagery data to predict metric such as the human-evaluated scenicness (Seresinhe et al 2015), economic value (Law et al 2019) or the sense of safety (Dubey et al 2016). A problem of these approaches is the lack of explainability of such methods in relating back to design which has an impact on usability. As such there is a need for greater research on causal inference in urban design which we will discuss in the last section. An example that provides such linkages is space syntax research (Hillier and Hanson 1989) which has a theoretical foundation that associate interpretable features of the built environment (spatial configuration) and performance metric such as pedestrian movement (Hillier et al 1993; Penn et al 1998) for evaluative urban design (Karimi 2012). A second approach is the use of agent based pedestrian simulation models (Turner and Penn 2002; Batty et al 2003; Crooks et al 2018), where agents are simulated within an environment following a set of rules over a period of time. These simulation models can create closed systems for scenario testing that can potentially isolate the effect of a particular design decision. The problem of these approaches is that such models would also be prone to omitted variable biases.


Challenges: data, causality and education

We have described in the last section, how data science methods can be used in the urban design process. We will conclude by describing a couple of potential challenges and consequently opportunities when applying data science methods in urban design.

A prominent challenge is the lack of labelled and open urban design process data. Despite the growing availability of large scale image data (Krizhevsky 2012) and architectural geometric data such as CAD (Koch et al 2019), there is still a large gap in the availability of urban design process data which is necessary to understand design mechanism or how design conceptualise. Recent research in Creativity and AI provides some inspirations going forward. For example, Google Quick Draw (Ha 2017) is a large scale sequential sketch database which can be used to train sketch completion or stroke prediction models. However, such sequential design process data is often not open source due to potential data privacy concerns and is often limited to simple design scenarios which differs from architectural design projects with multiple stakeholders. As such, this limitation represents an opportunity in the future to collect open urban design process data to better understand how design concepts emerge.

Another prominent challenge is the lack of interpretability with data science techniques such as machine learning models that have hindered its reliability to understand the causal design mechanism. For example, to what extent is there a causal connection between a specific urban design decision and a performance metric such as wellbeing. One perspective in looking at causal mechanism is through the domain of neuroarchitecture which studies the neurological processes of being situated in different space (Chatterjee et al 2021). In simpler terms, which parts of the brain is activated from observing different architectural scenes. Such research can reveal the biological mechanism in how design influences us.

A final challenge is the need to equip future generations of urban designer with the technical and analytical abilities to be able to creatively use modern data science techniques. An encouraging sign is that there are now postgraduate courses being taught on topics relating urban planning and design, computing and machine learning . We believe this trend will continue in the near future and will also be needed in undergraduate education in urban planning.

Summary

To end, this opinion piece provided some initial thoughts on how recent data science and machine learning methods can be applied in the urban design loop/process. This includes its integration on applying data mining methods for insight discovery at the inception of design, highly accurate reconstruction for augmented reality design, to the use of advance generative design methods for automated design synthesis and forecasting models to evaluate design. We have also highlighted a couple of challenges and thus opportunities on the

Presentation

Left Video Recording.

Conference Paper

Left Conference Paper.

Keywords

urban design, data science, machine learning, geographic data science, design computation.

Reference

DOI: https://doi.org/10.47330/DCIO.2021.DDJD8821

Bibliography

  • Batty, M., Desyllas, J., & Duxbury, E. (2003). Safety in numbers? Modelling crowds and designing control for the Notting Hill Carnival. Urban Studies, 40(8), 1573-1590.
  • Carmona, M. (2003). Public places urban spaces: The dimensions of urban design. Routledge.
  • Chaillou, S. (2020). ArchiGAN: Artificial Intelligence x Architecture. Achitectural Intelligence. Pp117-127.
  • Chatterjee, A., Coburn, A., & Weinberger, A. (2021). The neuroaesthetics of architectural spaces. Cognitive Processing, 1-6.
  • Crooks, A., Malleson, N., Manley, E., & Heppenstall, A. (2018). Agent-based modelling and geographical information systems: a practical primer. Sage.
  • Dubey, A., Naik, N., Parikh, D., Raskar, R., & Hidalgo, C. A. (2016, October). Deep learning the city: Quantifying urban perception at a global scale. In European conference on computer vision (pp. 196-212). Springer, Cham.
  • Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., ... & Bengio, Y. (2014). Generative adversarial nets. Advances in neural information processing systems, 27.
  • Guan, H., Yu, Y., Ji, Z., Li, J., & Zhang, Q. (2015). Deep learning-based tree classification using mobile LiDAR data. Remote Sensing Letters, 6(11), 864-873.
  • Fayyad, U., Piatetsky-Shapiro, G., & Smyth, P. (1996). From data mining to knowledge discovery in databases. AI magazine, 17(3), 37-37.
  • Ha, D., & Eck, D. (2017). A neural representation of sketch drawings. arXiv preprint arXiv:1704.03477.
  • Hansen, W. G. (1959). How accessibility shapes land use. Journal of the American Institute of planners, 25(2), 73-76.
  • Harris, R., Sleight, P., & Webber, R. (2005). Geodemographics, GIS and neighbourhood targeting (Vol. 7). John Wiley and Sons.
  • Hartmann, S., Weinmann, M., Wessel, R. and Klein, R. (2017). StreetGAN: Towards Road Network Synthesis with Generative Adversarial Networks. In proceedings of International Conference on Computer Graphics, Visualization and Computer Vision, 2017
  • Hartz-Karp, J., & Marinova, D. (2020). Using Deliberative Democracy for Better Urban Decision-Making through Integrative Thinking. Urban Science, 5(1), 3.
  • Hillier, B., & Hanson, J. (1989). The social logic of space. Cambridge university press.
  • Hillier, B., Penn, A., Hanson, J., Grajewski, T., & Xu, J. (1993). Natural movement: or, configuration and attraction in urban pedestrian movement. Environment and Planning B: planning and design, 20(1), 29-66.
  • Hillier, B., & Iida, S. (2005). Network and psychological effects in urban movement. In International Conference on Spatial Information Theory (pp. 475-490). Springer, Berlin, Heidelberg.
  • Hu, Y., Deng, C., & Zhou, Z. (2019). A semantic and sentiment analysis on online neighborhood reviews for understanding the perceptions of people toward their living environments. Annals of the American Association of Geographers, 109(4), 1052-1073.
  • Huang, Z., Heng, W., & Zhou, S. (2019). Learning to paint with model-based deep reinforcement learning. In Proceedings of the IEEE/CVF International Conference on Computer Vision (pp. 8709-8718).
  • Jumper, J., Evans, R., Pritzel, A. et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589 . https://doi.org/10.1038/s41586-021-03819-2
  • Karimi, K. (2012). A configurational approach to analytical urban design: ‘Space syntax’ methodology. URBAN DESIGN International 17(4)
  • Karras, T., Laine, S., Aittala, M., Hellsten, J., Lehtinen, J., & Aila, T. (2020). Analyzing and improving the image quality of stylegan. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 8110-8119).
  • Koch, S., Matveev, A., Jiang, Z., Williams, F., Artemov, A., Burnaev, E., Alexa, M., Zorin, D. and Panozzo, D., (2019). Abc: A big cad model dataset for geometric deep learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 9601-9611).
  • Koenig, R., Miao, Y., Achinger, A., Knecht, K., Konieva, K. (2019). Integrating urban analysis, generative design, and evolutionary optimization for solving urban design problems. Environment and Planning B: Urban Analytics and City Science
  • Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 25.
  • Law, S., Seresinhe, C.I., Shen, Y., Gutierrez-Roig, M.,(2020). Street-frontage-net: urban image classification using deep convolutionalneural networks.International Journal of Geographical Information Science,1–27
  • Law, S., Paige, B., & Russell, C. (2019). Take a look around: using street view and satellite images to estimate house prices. ACM Transactions on Intelligent Systems and Technology (TIST), 10(5), 1-19.
  • Li, X., Zhang, C.,Li, W.,Ricard, R.,Meng, Q.,Zhang, W. (2015). Assessing street-level urban greenery using Google Street View and a modified green view index. Urban Forestry and Urban Greening. Vol.14-3. pp.675-685.
  • Mildenhall, B., Srinivasan, P. P., Tancik, M., Barron, J. T., Ramamoorthi, R., & Ng, R. (2020, August). Nerf: Representing scenes as neural radiance fields for view synthesis. In European conference on computer vision (pp. 405-421). Springer, Cham.
  • Pettorelli, N., Vik, J. O., Mysterud, A., Gaillard, J. M., Tucker, C. J., & Stenseth, N. C. (2005). Using the satellite-derived NDVI to assess ecological responses to environmental change. Trends in ecology & evolution, 20(9), 503-510.
  • Penn, A., Hillier, B., Banister, D., & Xu, J. (1998). Configurational modelling of urban movement networks. Environment and Planning B: planning and design, 25(1), 59-84.
  • Polson, N.G. and Sokolov, V.O., (2017). Deep learning for short-term trafficflow prediction.Transportation Research Part C: Emerging Technologies, 79, 1–17. doi:10.1016/j.trc.2017.02.024.
  • Rangesh, A., & Trivedi, M. M. (2019). No blind spots: Full-surround multi-object tracking for autonomous vehicles using cameras and lidars. IEEE Transactions on Intelligent Vehicles, 4(4), 588-599.
  • Seresinhe, C. I., Preis, T., & Moat, H. S. (2015). Quantifying the impact of scenic environments on health. Scientific reports, 5(1), 1-9.
  • Silver, D., Huang, A., Maddison, C.J., Guez, A., Sifre, L., Van Den Driessche, G., Schrittwieser, J., Antonoglou, I., Panneershelvam, V., Lanctot, M. and Dieleman, S., 2016. Mastering the game of Go with deep neural networks and tree search. nature, 529(7587), pp.484-489.
  • Singleton, A. D., & Longley, P. (2015). The internal structure of Greater London: a comparison of national and regional geodemographic models. Geo: Geography and Environment, 2(1), 69-87.
  • Tian, Y., Suzuki, C., Clanuwat, T., Bober-Irizar, M., Lamb, A., & Kitamoto, A. (2020). Kaokore: A pre-modern japanese art facial expression dataset. arXiv preprint arXiv:2002.08595.
  • Tucker, C. J. (1979). Red and photographic infrared linear combinations for monitoring vegetation. Remote sensing of Environment, 8(2), 127-150.
  • Turner, A., & Penn, A. (2002). Encoding natural movement as an agent-based system: an investigation into human pedestrian behaviour in the built environment. Environment and planning B: Planning and Design, 29(4), 473-490.
  • Wartmann, F. M., Acheson, E., and Purves, R. S. (2018). Describing and comparing landscapes using tags, texts, and free lists: an interdisciplinary approach. International Journal of Geographical Information Science, pages 1–21.
  • Wortmann, T. (2017). Model-based Optimization for Architectural Design: Optimizing Daylight and Glare in Grasshopper. Technology|Architecture + Design.
  • Zanzotto, F. M. (2019). Human-in-the-loop artificial intelligence. Journal of Artificial Intelligence Research, 64, 243-252.
Retrieved from "http://wiki.designcomputation.org/home/index.php?title=Outlooks_and_Challenges_in_Urban_Design_X_Data_Science_Machine_in_the_Urban_Design_Loop&oldid=27492"