Here is a short excerpt from my chapter Simulation Tools for Social Performance – Immersive Building Simulations that talks about agent-based simulations as a tool to optimize a space’s social performance. Other simulation strategies that I write about include Tectonic Articulation, Social Space, Genetic Algorithms and Machine Learning. This short excerpt is for non-commercial internal, academic and research purposes only. Please do not copy or distribute. The entire text can be found in the book – please consider buying it.

Agent-based Occupancy Simulations

But also on a smaller scale, the analysis and prediction of spatial occupation patterns in various social spaces, most notably office environments, has moved more and more into the focus of contemporary design research recently, as a growing percentage of any developed country’s economy relies on intellectual capital rather than the means of production. Shifting away from the traditional Taylorist work culture with its long-established linear production logic, where the success of different spatial layouts could be easily measured (for example by simple counting the number of completed transactions, as routine assignments were handed down from one worker to the next), office tasks have become increasingly complex and knowledge driven. In Western European countries knowledge economy at this point represents about a third of all economic activities (Eurostat, 2013).

As work procedures became more flexible and spatial occupation patterns more loose and varied, and consequently could no longer be organized and measured according to Taylorist principles, new methods of assessing the performance of spaces need to be developed. Designers, such as the German Quickborner team with their Bürolandschaft concept, started in the 1970s to map the spatial relations between all agents in an office network in two-dimensional matrixes in order to find optimized office layouts. However, these design methodologies were still based on the clear assumption, that there is a linear relationship between the efficiency of a spatial layout and its successful work output.

In knowledge economies employees increase their respective radius of interaction due to the business’s intrinsically networked nature and various new types of knowledge work with their particular needs and mobility patterns emerge (Green and Myerson, 2011). The exchange of work is less vital than information interchange, communication, and human interaction, and the success of contemporary office environments increasingly relies on continuous formal and informal exchange of information and knowledge between actors in various different configurations. A space’s performativity largely depends on its capacity to spatially and semiologically frame the complexly interwoven interactions of its users.

In such a dynamic environment behavioral patterns are no longer linear but rather start to show emergent and unpredictable properties, which are the result of the repeated superimposition of (comparatively simple) interactions of its single constituent components (its agents), which gradually add up to the complex state of the emerging system. The result of such a non-linear process can no longer be predicted. This is also known as a “bottom-up” process, as opposed to a “top-down” process, in which the overall form is determined first. Such emergent systems however can be simulated by programming sets of simple agents that continuously interact with each other according to a basic set of rules. This was first achieved in 1987 by Craig Reynolds, who with his simulation program “Boids” successfully simulated the flocking behavior of birds. While such Agent Based Models (ABM) have been commonly used to simulate physical, chemical, biological, or sociological processes, architecture has only recently discovered their use for the simulation of life processes. Similar to a flock of birds or a school of fish, human crowds show complex non-linear behavior that is the result of a single agent’s behavioral rules or scripts that in turn are triggered by fellow agents or environmental features. As a consequence they constitute emergent systems, that can be successfully simulated by Agent Based Models (ABM). While plenty of commercial software programs, such as Cinema4D or Maya, offer readily available tools for crowd simulation, more complex life-like occupancy simulations at this point still require some scripting knowledge and the use of more specialized programs such as NetLogo (an open source program developed for an academic community) or Unity with its scripting extension (initially developed for the gaming industry).

Agent based design research is normally directed towards a concrete focal task in the form of a design research brief that acts as a experimental setup in relation to which empirical and statistical knowledge, conceptual and theoretical resources, simulation methodologies, and design ideas can be systematically brought together and explored.

In order to measure the different qualities and quantities of the agents’ interactions with each other, as well as with their environment, research focuses on the location, size, shape, configuration and features of informal office areas, such as communication zones, circulation areas and breakout spaces, where spontaneous communicative encounters and unscheduled opportunities for networking, collaboration and skill exchange most likely occur.

In order to set up a feasible first parametric model, the overall number of input parameters can be initially reduced within the conceptual framework of the simulation, gradually increasing complexity.

The main challenge is the development of a population of agents with plausible, life‐like individual behavioral rules that allow for the emergence of an overall plausible, life‐like collective event scenario. In order to accomplish this, any agent population needs to display two key properties of ‘life‐process modelling’. First of all, variable agent differentiation by status, affiliation, department, or position within the social network, implying behavioral difference, and secondly architectural frame‐dependency of behaviors, implying local selection from stacks of behaviors dependent on location and spatial architectural qualities.

The first property allows for the integration of client-specific employee structures and hierarchies, the second property facilitates a systematic semiological approach to the design, as it established relationships between the agents’ behavioral patterns and the furniture elements and features distributed across the space in question.

As the simulations run, systematically emulating various design proposals, relevant measures are recorded and stored to evaluate a scenario’s performance in terms of occupational patterns and communicative interaction. Those measures normally include location, frequency, relevancy and range of encounters and interactions as well as communicative histories, circulation patterns and occupancy maps (so called ‘heat maps”).

Data read out of a UNITY agent-based simulation setup. (c) Agent Based Semiology Research Group Vienna, Daniel Bolojan: 2018

Simulations’ results can not only be summarized and compared in order to judge the performance of a particular office layout. Once the collected data is statistically analyzed and set in relation to the quantitative parameters, such as position and number of specific reference points and furniture elements, throughout a series of related simulation scenarios (thus parametrically attaching it to the same spatial environmental features with changing locations), it can be used to train prediction algorithms in order to forecast agent behavior in novel and previously unsimulated scenarios.

Looking at statistical connections between the qualitative, i.e. semiological parameters of a spatial layout and its features, such as object design, environmental zoning or effective or affective conditions, and the behavioral patterns they generate might eventually lead to understanding the conception of office environments as an explicit design agenda to frame communication with a new and coherent system of architectural signification, rather than an intuitive participation in a historically evolving semiosis of the built environment, in order to develop an approach to architectural design that better engages with the opportunities and challenges of today’s networked society.

References:

Eurostat. (2013). Science, Technology and Innovation in Europe. Luxembourg: Publication Office of the European Union, p. 115.

Greene, C. and Myerson, J. (2011). Space for Thought: Designing for Knowledge Workers. In Facilities, Vol. 29 Issue: 1/2, pp.19-30, https://doi.org/10.1108/02632771111101304.

Agent-based Semiology for Simulation and Prediction of Contemporary Spatial Occupation Patters

Mathias Fuchs and Robert R. Neumayr

Abstract. Agent-based semiology is a powerful simulation and prediction environment for pedestrian simulation that allows for accurate balancing of complexity. Here, we describe a framework to simulate increasing behavioural interactivity between agents via agent-based modeling, together with a statistical approach to make the results amenable to a quantitative and automated analysis. That approach borrows ideas from crowd simulation and spatial statistics, notably fitting of Poisson processes, and computer graphics. The described process can simply be thought of as that of approaching an observed pattern by an overlay or additive mixture of grey-scale images each of which are distance transforms of physical objects. Thus, we describe the observed pattern in terms of interactions of spatial features which are akin to traditional BIM tags. We thus arrive at a remarkably concise prediction of the simulation outcome. The benefits of this simulation speedup is, on the one hand, to allow for higher optimization throughput, and on the other hand, to provide designers with quantitative feedback about the impact of their design on the simulation outcome.

Keywords: Parametric Modeling, Simulation, Agent-Based Semiology.

Introduction

Agent-based Semiology aims to investigate, analyze, simulate, and predict contemporary spatial occupation patterns in social spaces in order to understand and develop the performance criteria that interactively link these spaces, their interiors, and their users. The research ambition at hand is to develop a cross-disciplinary method of architectural design that generates spatial environments with high social performativity. 

Architecture channels its social processes through semiological associations as much as through physical separation and connection. It functions through its visual appearance, its legibility and its related capacity to frame its users’ communication. In that way the built environment is not just physically directing bodies, it is orienting socialized agents who have to understand and navigate complex spatial organisations, or  “As a communicative frame, a designed space is itself a premise for all communications that take place within its boundaries.” [21].

In a conventional design process, the designer tends to intuitively adapt and spontaneously intervene within the historically evolving semiological system of his respective cultural environment. The aim of agent-based semiology, however, is to move from this intuitive participation within an evolving semiosis towards a systematic and explicit design process, that understands contemporary spatial organisations as coherent systems of signification without relying on the familiar codes found in the existing built environment.

In today’s networked society and its communication-based working patterns, its increasingly open and dynamic environments, complex spaces and multi-layered social systems of use, the users’ behavioural patterns of interaction cease to be linear and simple to predict but rather start to show emergent and unpredictable properties. This emerging behavioural complexity is no longer a clear function of a number of static spatial occupation patterns, but the result of an iterative process in which the repeated superimposition of a set of comparatively simple interaction patterns of a system’s basic components (its agents) adds up to the complex state of the emergent system. Such nonlinear process, in which small-scale interactions govern a systems overall configuration, is called a “bottom-up” process, as opposed to a “top-down” process, in which the overall form is determined first in order to subsequently organize its constituent parts.

The results of these kind of processes can no longer be calculated or predicted and present serious challenges for predictive algorithms, suggesting the use of extensive computer simulations of its underlying dynamics. They can, however, be simulated using agent-based modeling which is commonly defined as “a computational method that enables a researcher [to] experiment with models composed of agents that interact within an environment.” [7]. In, 1987 Craig Reynolds was the first to successfully set up such a  simulation, reproducing the flocking behaviour of birds with this simulation program “Boids” [19]. Since then agent-based models (ABM) have been developed, refined and commonly used for simulations in the fields of physics, chemistry, biology, or social sciences, mapping the processes that we assume to exist in a real social environment [15].

While other tools and techniques for understanding social spaces, such as for example Space Syntax [10] have gained considerable popularity over the years, architecture has only recently discovered the use of agent-based models for the simulation of life processes. Similar to flocks of birds or schools of fish, human crowds show complex non-linear behaviour and thus constitute emergent systems, that can be successfully simulated by agent-based modeling. Therefore this research proposes their use for the simulation of architecture-related life processes. 

The analysis and prediction of emergent spatial occupation patterns in today’s social spaces, most notably contemporary office environments, has gathered more attention recently, as the economies of developed countries increasingly depend on the free flow of information rather than on the administration and exchange of goods and services. In Western European countries knowledge economy at this point represents about a third of all economic activities [6].

Moving away from long-established Taylorist office space layouts with its traditional linear logics of mono-directional workflow, where the success of different spatial layouts could be easily measured, work patterns in today’s knowledge economy has become increasingly complex, flexible, and interwoven and can therefore no longer be organized and evaluated according to Taylorist principles. Consequently, new methods of assessing the performance of spatial layouts need to be developed.

Designers such as the German Quickborner team with their Bürolandschaft concept, started already in the 1970s to develop novel office layouts that were essentially real-life diagram-spaces based on the optimization of spatial relations between groups of employees. While this design methodology was still relying on the assumption, that there is a linear relationship between a spatial layout’s organisational efficiency and its work output, one of its major innovation was that it focused on the patterns of communication rather than its contents, thus establishing the flow of information as a generative tool. The other key innovation was the abolishment of spatial hierarchies in order to foster informal communication which was considered critical in a cybernetic organisational model [14], or as Quickborner’s Ottomar Gottschalk put it: “Informal conversations are not only useful – in all likelihood  they are actually crucial.” [8].

In knowledge economies employees increase their respective radius of interaction due to the  intrinsically networked nature of contemporary workflow and as a consequence various new types of knowledge work with their particular needs and mobility patterns emerge [9]. The sharing of work, goods or products is less important than information interchange, communication, and human interaction. A space’s performativity therefore largely hinges on its capacity to spatially and semiologically frame the constant informal and formal transfer of knowledge between its users in various different configurations.

Therefore, a design brief of a contemporary office environment acts as an experimental setup in which empirical and statistical knowledge, simulation methodologies, and design ideas are systematically brought together. In order to measure the various emergent patterns that arise from the agents’ continuous interactions with each other, as well as with their environment, the research focuses on the office space’s breakout space, which is its most informal area, where spontaneous communicative encounters and unscheduled opportunities for networking, collaboration and skill exchange can easily occur. At the same time the space’s furniture elements also allow for planned or organized meetings or conferences of various sizes and configurations.

Simulation methods and ACL – scenario matrix 

The research sets up and refines agent based life process simulations, in which the semiological code is defined in terms of the agents’ behavioral rules or scripts being triggered by specific environmental features as well as by the interaction with other agents. The simulation is run in two independent programs in parallel, NetLogo and Unity, with the same setup and variable values, in order to be able to compare results and data.

NetLogo (see Fig. 1) is a program designed for agent-based simulations with built-in processes that already solve typical agent based simulation scripting problems. As it is purely code based it is fast, scalable and data extraction is easy.

Fig. 1. The Netlogo simulation environment.

Unity is a component based software program, that is very popular amongst game developers used to develop multiplayer video games across various platforms. Its complexity allows for sophisticated agent interactions and can also be fully script driven. 

In order to understand, simulate and predict comprehensive life-like office scenarios with complex behavioral patterns within a controlled simulation, a clearly defined small-scale office breakout space is used to script and test relationships between agents and the environment. The task is the development of a population of agents with life‐like individual behavioral rules that allow for the emergence of a simplified, yet overall plausible collective event scenario. For  this, any agent population needs to display two key properties of ‘life‐process modeling’. 

First of all, there needs to be agent differentiation by status, affiliation, or position within the social network, implying behavioral difference as agents interact with each other. Real-life social interactions are complex in nature, as they depend on multiple variables, that have to be effectively weighted to be made operational in a simulation. While this research has been looking into using social network analysis to extract relevant social network features (for example from email databases) to later inform agent-based life process modeling, at this point  agents’ behaviour, such as probability and duration of interaction, is driven by randomly assigned accumulative interaction values. Agent will choose to interact with the agent agent, who has the highest interaction value within a fixed range of values that who is available (i.e. not occupied) within a defined distance.

Secondly there needs to be architectural frame‐dependency, implying different behavioral patterns depending on location and spatial architectural qualities. To that end, architectural features, which are typical for an office break out space, such as different types of tables, a reception desk, or a coffee machine, are added to the simulation environment, once the basic logic is set up. These features are assigned interaction values similar to the agents, that govern the agents’ interaction with them.

Like in other strands of design research, continuously refining the digital processes becomes an important issue once a basic logic is established[16]. Therefore the complexity of the simulation is increased continuously, step by step. While the simplest movement pattern in spatial simulation is the random walk [17], the initial simulation setup is an agent walking around the scene unaware of his surroundings. 

Simulations gradually increase in complexity and for better systematic comparison are organized in a 2-dimensional matrix system. On the vertical axis we define the agent complexity (agent capacity level – ACL), starting from the simplest possible agent as described above (ACL 1.0). In each subsequent step, the agents’ capabilities are systematically extended (i.e. collision detection, agent interaction, object interaction, etc.), up to a simplified office setup that allows for agent to agent interaction as well as for an interaction with a number of common furniture elements, such as a reception desk, a coffee machine, high tables, low tables, and a meeting table. The result is an accumulative build up of potential agent faculties that allows for direct comparison of individual ACLs and therefor speculation on possible success criteria and relevance of agent capacities. 

On the horizontal axis, each capacity level is tested in four parallel simulation scenarios (A, B, C, D) in order to produce a more robust data set. These scenario differ slightly from each other in terms of their floor pan layout, the location of doors, and the position of interaction objects. The parameters that remain constant are a maximum number of 16 agents per simulation and the total run time of 30 min. During runtime relevant data, such as the agent’s positions, their encounters and interactions, is constantly collected from the simulation and stored for later analysis. For ease of comparison the data is used to create a number of visual quantifiers, such as heat map showing the concentration of occupancy (density) and trail maps tracking the movement of each individual agent. 

The collected simulation data from the first three simulations (A, B, and C) is used to train a prediction algorithm that finally is tested against the empirical data set of the last simulation (D) for accuracy, where the prediction algorithm is confronted with a novel scenario condition.

Statistical description and prediction

Overview and goal

Gaining an approach to a statistical description is necessary for bridging the gap between pedestrian simulations on one side, and 3D modeling. We aim at automating the knowledge transfer in an objective and repeatable way, by deriving a sufficiently general model of the interaction between people and their environment. We emphasise a quantitative, interpretable, easily trained algorithm inspired from statistical learning, that is adaptive and easily extrapolated to new scenarios, in order to leverage the potential of pedestrian simulation for real-world design.

Spatial data analysis [4] deals with regressing observed movement patterns on spatial features. Whereas in a usual linear regression model, the values taken by a single dependent variable are explained by relating them to a linear combination of the independent variables, the so-called features, in a spatial data analysis task an entire spatial field of observations is related to a linear combination of independent feature fields. For instance, the dependent field could be that of successful oil rigs, and the independent fields could be the maps of geographical and geological characteristics. Here, the dependent field is given by the observed pedestrian movement pattern, and the independent feature fields are given by explanatory physical or architectural features of the space. Then, observations of a linear combination of the dependent variables blurred with Gaussian noise. 

Figures 2 and 3 demonstrate the  inspiration visually: to bridge the gap between hard geometrical features and smooth pedestrian behaviours.

Fig. 2. The discrepancy between “hard” geometry and “soft” pedestrian movement patterns can be bridged tentatively by assuming the existence of a guiding principle that steers the pedestrian density according to the proximity to the geometrical pattern. In this work, we are formalising the idea of describing the observed pedestrian density as an overlay of proximities to hard geometrical or physical features of the space.

Fig. 3. A spatial pattern is assumed to be generated by input fields, generated by objects. Typically, doors and tables are the “seed crystals” of these fields.

Development of a statistical model

Therefore, we need to find a way to describe regressors or dependent variables. We used a notion of one-channel image, reflected in a C++ class, to capture the spatial impact of a feature in the form of its distance image [12]. A feature is a user-supplied “explanatory” function. Here, we use the distance fields from objects but in principle any kind of one-channel image can be used for the task.  Such a function can be thought of as an gray-scale image, overlaid onto the 2d rectangle representing the scene’s bounding box in plane view. A typical function could represent the distance to a point of interest such as the central table in a meeting room. Thought of as an image, it consists of radial isolines around the table, getting darker and darker farther away from it. In a first step, one models the impacts of the most apparent spatial features, acting as attractors: doors, tables, toilets, staircases, maybe windows, etc. Their impact is modeled by a function which decreases steadily with higher distance to the object. In general, there is a variety of possible “influence maps”.

In concrete statistical terms, estimating the weights or contributions of each feature is done by fitting a Point Poisson Model to the Data with the Maximum Pseudolikelihood Method [3 and 9]. Fig. 4 shows the central object of study: the pedestrian density represented at as a heatmap. Concretely, the process can be thought of as first computing the empirically observed heatmap and then interpreting it as an overlay as pictured in Fig. 6 of distance images to these spatial features. 

Fig. 4. The building block of statistical reinterpretation of the observed pedestrian occupation pattern is the heatmap. A small number of these is used to approach the observed one. This figure illustrates the complexity inherent in an observed heatmap. Moreover, the dependence of the visual appearance of the heatmap on the bandwidth used for its computation, is often underestimated. This phenomenon corresponds to the ubiquitous property of natural patterns to exhibit structure on every scale from small to large whereas artificial patterns are often constrained to fewer scales.

Attaching features to physical objects and statistical model fitting

The aim of statistical analysis is twofold: fostering the understanding of the spatial process by descriptive processing, and the extrapolation of the learned model to new scenarios in a prediction. In this second step, we derive a way to draw the density of pedestrian activity, as generated in a simulation, onto the design canvas in an adaptive manner. We accomplish that goal by employing techniques and algorithms from point processes. In particular, we show how rudimentary BIM (building information model) tag information gives rise canonically to a set of spatial features. We then go on to describe how a Poisson model fitting gives rise to a statistical model which associates to a design intervention a smooth (“parametric”) adaptation of the predicted pedestrian density. See [20] for an overview of density estimation and its relationship with Gaussian blurring, and the subsection below for the exact description of the fitting method of the features. Fig. 4 shows the central object of study: the pedestrian density represented at as a heatmap.

We validate the generated predictive model by careful separation of the data into disjoint learning and testing sets. A particular benefit of the method is its susceptibility to produce not just predictions — as, for instance, a neural network would — but instead to generate an understanding of how the model works, by yielding concrete and interpretable coefficients associated with the spatial features.See Fig. 7 for an example of prediction results on ACL 4.4.

Fig. 5. The concept of reinterpreting and observed pattern in terms of an overlay of simpler ones is quite powerful. Here, we show how any uniform random point pattern, i.e. a noise pattern, can be described almost perfectly as an overlay of generic heatmaps – sine waves in this case. The approximation principle is closely related to consequences of the theory of Fourier series.

Fig. 6. Each spatial object generates its distance transform. Here, six types of objects – door, window, three standing tables and a desk, each generate a distance function. Finally, these are overlaid additively to generate a statistical re-interpretation of an observed pedestrian pattern. The process of fitting the parameters of a Poisson process described here is the one that determines the contributions or weights of each single such feature image to the overlay.

Fig. 7. Result of the prediction. Learning sets are A to B, true observed pattern is D.

Description of the fitting algorithm

The actual core of the method is the computation of the weights of each single image — typically, a distance image. Each weight gives the image’s relative contribution to the overlay which is intended to be describe the output pattern given by the observed pedestrian locations, as closely as possible. The relevant sub-discipline of statistics is that of spatial data [3], and in particular that of spatial point patterns [5]. To simplify matters, a spatial point pattern can, in that context, just be thought of as a finite collection of two-dimensional points. We use the R system of statistical programming [18], and in particular the popular package “spatstat” for spatial statistics [2].

The translation of our geometrical problem into the language of spatial statistics is very straightforward: The locations immediately define a spatial point process, and each distance image defines a spatial function and thereby a covariate (or independent variable as in the more classical context of usual univariate regression). We interpret the latter locations as a spatial point process and  these weights by fitting a point process model to the observed point pattern with the “ppm” method of the spatstat R package. This function, in turn estimates the weights or contributions of each feature by fitting a Point Poisson Model to the Data with the Maximum Pseudolikelihood Method [1 and 11].The function yields a fitted model, and each covariate’s coefficient is interpreted as the desired weight. Note that this method is superior to the naive way of performing a pixel-wise linear regression because it takes the evident correlations between pixels into account. We use the “Poisson” interaction model [13] of the ppm function which is a simplification because it is the simplest model and  already yields meaningful results. Additionally, we integrate a slight non-linearity into the fitting process by taking interaction terms up to cubic order into account.

Generalisability

It should be noted the goal of the prediction is not simply to be as accurate as possible. Indeed, greater accuracy can always be achieved at the expense of generalisability: Learning more about a particular constrained design task comes at the cost of giving up the applicability to a more general design language. On the other hand, a general learned model can not adapt to the particularities of a special situation. 

Here, we have chosen to confine the study to a particular space, so that we have been leaning on the side of accuracy. See Fig. 5 for a demonstration of an arbitrarily good fit of the given data. Input fields are sine waves, and  the overlay or superposition corresponds to the terms of a Fourier series converging against a limit in Hilbert space. Here, we generated a random spatial pattern and have demonstrated extreme “overfitting” by approaching that pattern to a stunning degree, taking a large number of terms into account. On the other hand that illustration shows the power and flexibility of the simple overlay approach. Therefore, in an architectural context, the question of overfitting remains salient, imposing a strong trade-off between prediction accuracy and generalisability.

Limitations and drawbacks

While the simulation’s taxonomy, scripts and results as well as the developed prediction algorithms yield very promising results, the office space layout initially chosen for the simulation setup starts to show certain restriction and limitations. Although size and program of the simulation setup were intentionally kept quite confined in the beginning in order to better control the overall simulation and its results, later on this decision proved to be increasingly limiting.

The prediction’s accuracy increases with each additional object and feature included in the simulation, however, the small scale of the simulation space sets a clear limit to the overall number of one-channel images that can be meaningfully implemented. Not only does the intended use as an office breakout space restrict the types of possible objects and features, the size of the chosen office simulation space also sets a physical limit to maximum number of objects such a space can purposively hold. The next stage of this research will therefore seek to expand the scope of the simulation space, adding an additional office program as well as more one-channel images.

Regarding the geometric-statistical method, a possible drawback is that it does not apply immediately to optimization because on the one hand it only predicts densities which are by definition relative (their integral is one). Another drawback is that it is unclear how to determine which classes of layouts a prediction generalizes to. Thus, it would be desirable to identify either geometric, architectural or maybe even socio-cultural traits of a space which should be treated as being within the prediction scope of the algorithm.

Summary

We have described a unique approach borrowing insights and techniques from architecture, agent-based modeling, statistics of point processes, and computer graphics into a novel framework for semiological interpretation and prediction of contemporary spatial occupation patterns. The methodology introduced in this research is tailored towards measuring and simulating the social phenomena that stem from our increasingly complex built environment.

In total, we have shown that a simple process of overlaying spatial features corresponds to the statistical process of fitting the parameters of a Poisson process.

Outlook

The proposed statistical method can be described as “quantitative semiology” and has, as such, a vast field of possible applications in architecture and urbanism.

The first logical step consists in applying an optimization engine like for instance an evolutionary algorithm to exploit the prediction for “machine design”: one extracts from the predicted density a success measure linked to the functionality of the space or the architecturally desired functional outcome.  It could be defined by the amount of conversations, or by their comprehensiveness, or by some measure of the receptiveness of the design, etc. Then, a density prediction leads to a prediction of that particular chosen measure. From that point on, it lends itself to consider a larger number of possible designs or even an entire parametrized family, and to relate the outcome – the success measure – directly to these parameters. There are straightforward parameters such as sizes, distances and other quantities of geometric provenience, as well as well-studied design parameters coming, for instance, from the space syntax grammar [10]. In general, however, it becomes necessary for actual optimisation to assign some “space DNA” quantities to each proposed design. The more insightful these parameters are chosen, the more likely is the success of a machine learning strategy for associating the social functionality. Hence, one will be in a position to ask for prediction and optimization of the success of an entire design family. For this, it would be helpful to have better design measures and fitness criteria. In particular, the question about insightful and parsimonious design parameters that are suited to describe a design’s functionality remains to be answered in an interdisciplinary approach.

There is an ample spectrum of possible future research possibilities. In particular, it will be worthwhile to exploit exploitation of the more realistic 3d features of a Unity simulation, and pose the question how the agents’ behavior is influenced by their field of vision, possibly to be described in combination with an evolutionary algorithm. In the end, one will learn much more about the complexity of pedestrian behavioral patterns and their interaction with the surrounding space.

References

[1] Berman, M. and Turner, T.R. Approximating point process likelihoods with GLIM. Applied Statistics 41 31–38 (1992).

[2] Baddeley, A., Rubak, E. and Turner, R. (2015). Spatial Point Patterns: Methodology and Applications with R. Chapman and Hall/CRC Press (2015)

[3] Cressie, N., Statistics for spatial data. John Wiley and Sons, Inc (1993).

[4] Daley, D.J., Vere-Jones: An Introduction to the Theory of Point Processes. Volume I: Elementary Theory and Methods. Springer (2003).

[5] Diggle, P.J.: Statistical Analysis of Spatial Point Patterns. Arnold, London, second edition (2003)

[6] Eurostat: Science,Technology and Innovation in Europe. Luxembourg: Publication Office of the European Union, p. 115. (2013).

[7] Gilbert, N.: Agent-Based Models. Thousand Oaks: Sage Publications Inc., p. 2. (2008).

[8] Gottschalk, O., Architekt of the Quickborner Teams, Symposium “Bürolandschaft” at the documenta 12 in Kassel. (2007).

[9] Greene, C. and Myerson, J.: Space for Thought: Designing for Knowledge Workers. In Facilities, Vol. 29 Issue: 1/2, pp.19-30, (2011).

[10] Hillier, B. and Hanson, J.: The Social Logic of Space. Cambridge: Cambridge University Press. (2003).

[11] Jensen, J.L. and Moeller, M.: Pseudolikelihood for exponential familymodels of spatial point processes. Annals of Applied Probability 1 445–461 (1991).

[12] R. Kimmel and A.M. Bruckstein, “Distance maps and weighted distance transforms,” in Proceedings SPIE-Geometric Methods in Computer Vision II, San Diego (1996).

[13] Kingman, J. F. C.: Poisson Processes. Clarendon Press (1992)

[14] Kockelkorn, A.: Bürolandschaft – eine vergessene Reformstrategie der deutschen Nachkriegsmoderne. In ARCH+ Zeitschrift für Architektur und Städtebau, Vol. 186/187 (2008).

[15] Macy, M. and Willner, R.: From factors to actors: Computational sociology and agent-based modeling. In Annual Review of Sociology, 28, pp. 143-166. (2002)

[16] Neumayr, R. and Budig, M.: Generative Processes – Script Based Design Research in Contemporary Teaching Practice. In Paoletti, I. (Ed.), Innovative Design and Construction Technologies. Milano: Maggioli S.p.A., p. 172. (2009).

[17] O’Sullivan, D. and Perry, G.: Spatial Simulation. Exploring Pattern and Process. Chichester: Wiley-Blackwell, pp. 97-131. (2013).

[18] R Core Team: R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna.

[19] Reynolds, C.: Flocks, herds, and schools: A distributed behavioral model. In: Proceedings of the 14th Annual Conference on Computer Graphics and Interactive Techniques ACM. 21 (4), pp. 25–34. (1987).

[20] Silverman, B.W.: Density Estimation for Statistics and Data Analysis, In: Monographs on Statistics and Applied Probability, London (1986).

[21] Schumacher, P.: Advanced Social Functionality Via Agent-Based Parametric Semiology. In: Schumacher, P. (Ed.), Parametricism 2.0. AD 02/2016. London: Wiley, p. 110 (2016).

ion luca. 2018. 400 x 400 mm. c-print on photographic paper. edition 2 + 1 AP. price upon request

]]> http://www.unsquare.at/?feed=rss2&p=1343 0 http://www.unsquare.at/?p=1314 http://www.unsquare.at/?p=1314#comments Sun, 29 Apr 2018 17:41:30 +0000 0801 http://www.unsquare.at/?p=1314 our current research project Agent Based Parametric Semiology is part of the traveling exhibition understanding art and research, whose first stop will be at the dunedin school of art in new zealand. thanks to the angewandte and the FWF for the support.

tresorstiege. 2018. 400 x 400 mm. c-print on photographic paper. edition 2 + 1 AP. price upon request

The image you just clicked on shows the project VERTICAL ABYSS by students Ke Liu, Angeliki Tzifa and Dongliang Li.

The Best Building Money Can Buy – The Future of the Skyscraper Outside its Current Capitalist Logic

“Once you learn to look at architecture not merely as an art more or less well or more or less badly done, but as a social manifestation, the critical eye becomes clairvoyant.” – Louis Sullivan

In 1896 Dankmar Adler and Louis Sullivan completed what would eventually be their practice’s last mayor built project, the Guaranty Building in downtown Buffalo. At the time of its completion however, the building had already changed its name. Carefully located at the intersection of Church and Pearl Street close to most of the city’s and the county’s official buildings, it was initially supposed to be called The Taylor Building, named after a local businessman by the name of Hascal T. Taylor, who had developed the early skyscraper as a speculative office building in the wake of the construction boom surrounding Buffalo’s quick but temporary rise to importance as its former mayor Grover Cleveland was elected 22^nd president of the United States.

It was only due to the entrepreneur’s untimely death before the structure’s completion that the contracted Guaranty Construction Company decided to continue with the project alone. But also the architectural firm Adler & Sullivan met their fate. Ironically enough, a continuous decline in commissions, resulting from a severe recession, known as the Panic of 1893, that in turn started with the burst of a speculative bubble in Argentina, forced the two architects to dissolve their partnership in 1894.

So, speculation, I would argue, has from the early days on been one of the key drivers for real estate development in general but also for its most prominent protagonist in particular: the urban skyscraper. However, little attention has been given to the question of how the principles of financial speculation have affected its historic and typological development.

In her book Form Follows Finance, Carol Willis for example argues that high-rise buildings, especially in Chicago and New York, have always been speculative developments, and that their form, location, and distribution throughout the city are the result of complex interactions of parameters such as plot sizes, local or regional building patterns, cost effective construction technologies, fluctuating real estate cycles, building codes and zoning laws [1].

Christopher Marcinkoski defines speculative urbanization as “[...] the construction of urban infrastructure or settlement for political or economic purposes, rather than to meet real (as opposed to artificially exaggerated) demographic or market demand.” [2] Giving an overview of notable past and present speculation bubbles, he points out a fundamental change in what drives contemporary urban development. Urbanization is no longer a response to economic growth but is rather deployed as a driver for economic growth, thus becoming a preferred instrument of economic production.

As the amount of capital that is channeled towards real estate increases, the degree to which space functions as an asset has increased radically and the large-scale effects, that high-rise buildings as objects of financial investments have on their immediate built environment become visibly more clearly. It seems that cities nowadays have become complexly multi-layered commercial environments, where space is being privatized in all three dimensions and all property—similar to the stock market—is subject to fluctuating value cycles. Buildings have become assets and air rights have developed into a speculative commodity.

In that sense, for some time now, New York, and more specifically Manhattan with its long history of converting incoming capital to high dense building mass has been on the forefront of this development and therefore seems to be uniquely equipped as a location to examine the phenomena of speculation, asset architecture and its effects.

In direct contradiction to the standard evaluation criteria typically associated with any building in an urban context, the contemporary pencil shaped condominium towers, that sprout all over the city, only operate as financial investment assets. Following the national trend, speculative vacancy in New York has grown rapidly to 12% of the housing market [3] and explains why so many facade windows of upscale Fifth, Madison and Park Avenue apartment buildings remain dark every evening. In 2004 across the United States 23 percent of all the houses acquired were for investment purposes and another 13% were obtained as second homes [4].

As of now, New York is still the third most expensive city for prime real estate in the world. According to Knight Frank Research [5], in December 2016 one million US$ let you buy 26 m2 of prime property. At that time only Monaco and Hong Kong did better. For the amount of one million US$ you could afford 20 m2 of luxurious property, in the exclusive Mediterranean princedom that same amount of money bought you a meagre 17 m2 of first class real estate. But cities in the Western hemisphere appear to be on the decline, and Asia seems to be on the rise. In 2016 luxury residential market performance in New York was at 3.50% and market value is forecast to not increase at all in 2017. In comparison, the four top performing cities in the Prime International Residential Index (PIRI) 2016 were all Asian, Shanghai coming in first, closely followed by Beijing and Guangzhou (all three of them with an increase of prime real estate value of around 27%) and Seoul. And contrary to New York, Shanghai’s top real estate segment is predicted to increase in value by another 8% in 2017 [6].

It was only recently, and after looking back at a considerable record of burst real estate bubbles, that the notion of financial speculation as the main driver for urban tower development has started to rapidly gain influence within the contemporary discourse about high-rise structures within an urban context, considerably shifting the way we look at the history of the skyscraper. More than any other contemporary building typology, the tower has always been widely understood, by the general public as well as within the profession, as a symbol for innovation, modernity, technological advancements, and progress. As a consequence, historically, two narratives dominated the discourse about the development of the skyscraper. The first account is chronological, understanding the tower as a in itself shapeless typology that adapts throughout history to the prevailing styles, ideas and agendas of any given specific time period. The second reading focuses on technological aspects, linking constant advances in building and construction technology as the main drivers behind skyscraper development. However as both of these narratives can only operate post festum, they lack the projective mechanisms that would be needed to allow for productive speculations about this dynamic typology’s future.

But now, as these narratives are being cast aside by an increasing criticism of global capitalism, its relation to speculative capital, global wealth distribution and its impact on our built environment, high-rise buildings are rediscovered as ideological objects to bring back a strong social and political agenda to architecture, tying in with modernism’s long lost social project, where rationalization, new technologies, automated production techniques and innovative materials were supposed to provide affordable high-quality products and solve social problems on a global scale.

So, while the contemporary architectural production, such as Norman Foster’s 700-foot ultra-thin high-tech residential tower on 610 Lexington Avenue, just behind Mies van der Rohe’s Seagram Building, is still well embedded within the current architectural paradigm and its speculative logic, shifting the architectural discourse back towards a more socially biased agenda has also jump-started academic speculation about the socially responsible future of the skyscraper, following Zaha’s famous dictum that “[w]ithout experimentation not much can be discovered. […] I think there should be no end to experimentation.” [7]. And in its more interesting recent moments, the results of these experimental design approaches manage to be innovative on multiple levels.

“New York Horizon”, the winning entry of the 2016 Evolo Scyscraper Competition [8] for example explores a new high-rise typology by excavating Central Park down to its bedrock, thus creating a seven mile long and 1000 feet tall wall of skyscrapers along its perimeter, with an unobstructed view onto the newly emerging underground landscape.

Referring to one of Central Park’s designers, Frederick L. Olmsted’s initial intention, to provide a central common green space, equally accessible to all citizens, the project clearly sets its social agenda in providing an additional seven square miles of inhabitable indoor space with direct view to the park scenery, a commodity that has become scarce in recent years as new towers around Central Park have continued to rise higher than ever before as have the prizes of property with a direct view to it.

By doing so, at the same time, the project inverts the modernist understanding of the relationship between building and landscape, making the purposefully faceless architecture the mere background for the natural landscape as a centerpiece.

But, just like the “Manhattan Tower”, a twelve kilometer high extrusion of the “site formerly known as Central Park, scraping the Stratosphere” in Jimenez Lai’s dystopian graphic novel Babel [9], the proposal also makes us reconsider our understanding of scale and the reciprocity between volume and void. It reveals the complex interplay between the height limits of large-scale human structures and the immediate, yet not directly graspable dimensions of the geological and geographical structures that surround us and it reminds us of the unearthed potentials of the contemporary urban fabric’s inherently stratified nature and vertical complexity.

This year’s AAD PPD students responded to the design brief In equally innovative ways, speculating about how novel ideas about asset driven architecture could start to shape the program, structure, materiality, form and visual nature of the contemporary city tower.

Responding to the prevailing speculative urban high-rise development, the studio’s thesis is how to design a Manhattan skyscraper that, rather than fighting financial investment logic, subversively operates within the capitalist paradigm in order to come up with building proposals that will not only generate a return on investment that outperforms conventional investment strategies, but that will also at the same time be able to operate as a social, cultural or programmatic condenser within the surrounding city fabric.

Understanding the mechanics of asset architecture as a complex social phenomenon rather than an abstract and detached market mechanism, students are able to conceptualize the given challenge simultaneously on a social, cultural, programmatic and aesthetic level.

A series of seven sites in midtown Manhattan along the south end of Central Park that form part of Billionaires Row, becomes the testing ground for this semester’s speculative design research.

Some teams investigate alternative programs that might lend themselves as possible investments and their potential to be organized in new vertical ways.

Starting from the observation that burial space has become a scarce commodity in New York, leaving only one cemetery still selling space for remains on the crowded island of Manhattan, the project “Death in Manhattan” by students Carrie Frattali, Jeon Bosung, and Zhao Xiaoyu for example investigates a novel type of real estate by turning burial space into a commodity and developing it as a vertical program.

The vertically structured mausoleum can produce a huge variety of burial typologies at varying price levels, depending on prominence, orientation, views or relativity to ceremonial or public spaces. Vertically densely stacked mausoleums, niches, stacks of urns or caskets are contained within the structural elements of the building and connected by a procession of programs. The tower’s novel typology rotates the traditional ceremonial progression into verticality, thus elevating the process of ritual, meditation, and contemplation in a vertical version of the traditional landscaped cemeteries. The towers facade consists of a series of louvers that change direction and depth, highlighting program changes on the interior as well as providing a variety of light quality throughout the tower.

Another important field of research appears to be the investigation into new residential typologies. With new condominium towers going up across the city nowadays in the form of super slim pencil towers, the floor plans of New York’s most expensive apartments have locked into one standardized one storey layouts that wrap around a massive central core, increasingly failing to offer differentiated experiences in a saturated market.

“Vertical Abyss”, a project by students Li Dongliang, Liu Ke, and Angelika Tzifa sets out to redefine the concept of luxury housing by developing an interiority of accentuated verticality, that reflects the soaring upright directionality of the tall building’s overall shape. 136 apartments units are devised with minimized footprints on multiple levels, which are connected internally by a moving platform. They are stacked and arranged following a vertical packing algorithm to interlock around a series of vertical void spaces that act as atria for the respective units.

Other design projects finally call into question the well-established financing models for contemporary asset architecture. “Ten Million New York” by students Chen Xiaonan, Wu Mengyue, and Zhong Jianbao for example suggest a participatory bottom-up real estate crowd funding model in which a multitude of small private investors contribute a modest investment to the overall cost of the skyscraper. In return they do not only receive tradeable shares of the building but also gain the exclusive right to access the tower and its semi-private facilities which are being shared by all the investors. The building’s interiority in turn consists of a vastly different spaces, levels, plateaus, niches, and crevices generating a multiplicity of atmospheres with their respective effective and affective conditions to be temporarily inhabited by its shareholders.

Consequentially the continuously variegated outdoor space that entwines the building, reads as a complex contemporary three-dimensional interpretation of New York’s private green spaces, such as Gramercy Park.

Endnotes:

[1] Carol Willis. Form Follows Finance. Skyscrapers and Skylines in New York and Chicago. (New York: Princeton Architectural Press, 1995)

[2] Christopher Marcinkoski. The City That Never Was. (New York: Princeton Architectural Press, 2015)

[3] Sam Roberts, “Homes Dark and Lifeless, Kept by Out-of-Towners.” The New York Times, 06 July 2011. Online Edition.

[4] “In Come the Waves” Economist, 06 November 2008. http://www.economist.com/node/12501011.

[5] Kate Everett-Allen, “Going up, going down.” The Wealth Report 2017. ed. Andrew Shirley. (London: KnightFrank, 2017).

[6] James Roberts, “Future view.” The Wealth Report 2017. ed. Andrew Shirley. (London: KnightFrank, 2017).

[7] Hans Ulrich Obrist, “Foreword” Zaha Hadid Early Paintings and Drawings. (London: Serpentine Galleries and Koenig Books, 2016)

[8] “New York Horizons” Evolo Skyscraper Competition 2016. http://www.evolo.us/competition/new-york-horizon/.

[9] Jimenez Lai, “Babel” Citizens of No Place. An Architectural Graphic Novel. (New York: Princeton Architectural Press, 2012.