adaptive distributed architectural systems

446

ACADIA 2012 // Synthetic Digital Ecologies

will soon become a regular element of the built environment.

of architecture students and practitioners to adaptive design techniques, adaptive architectures

architectural design. We argue that with presently available technology and an increased exposure

created during workshops based on the theme of integrating adaptive distributed systems into

opens new aesthetic possibilities for designers. We also present examples of student work

as dynamic systems with the ability to adapt to the fluctuating environments in which they exist

building, play an important role in the aesthetic result of a design. Therefore, conceiving buildings

daylight distribution. In addition, building envelopes, being the most publicly visible part of a

issues of distributed shading, power generation through integrated thin film photovoltaics, and

explore the negotiative potential of adaptive systems in architecture. The ASE prototype addresses

of energy performance and occupant comfort and thereby offers an ideal scenario in which to

envelope (ASE). The building envelope was chosen because it is a common place to address issues

application of adaptive systems to architectural design through the prototyping of an adaptive solar

history, we clarify the term adaptive and its use within the field. This allows us to explore the

Artificial intelligence has a long and rich history in the field of architecture. Building upon this

ABSTRACT

Architecture and Sustainable Building Technologies, Department of Architecture, ETH Zurich

Arno Schlueter

Architecture and Sustainable Building Technologies, Department of Architecture, ETH Zurich

Zoltan Nagy

Architecture and Sustainable Building Technologies, Department of Architecture, ETH Zurich

Dino Rossi

ADAPTIVE DISTRIBUTED ARCHITECTURAL SYSTEMS

// Synthetic Optimizations

447

448

ACADIA 2012 // Synthetic Digital Ecologies

1

INTRODUCTION

of an adaptive control system. Note that the responsiveness is still there, but in addition, adaptive control allows the automatic reprogramming of the responsiveness. We will employ this notion of adaptive for the remainder of this paper.

Negroponte’s SEEK project (1970) had physically explored the idea of a self-reconfigurable architecture,

but on a smaller scale than that proposed by the Generator project. SEEK combined a three-axis CNC

thereby adapt their behaviors to the desires of the occupant(s) over time. A system would be free to optimize itself for energy performance when the space is unoccupied, or could even nudge an occupant toward “greener” habits by testing the boundaries of what the occupant is willing to accept as comfortable. In essence, it would be an ongoing dialogue between occupant(s) and building, navigating their sometimes shared and sometimes contradictory desires. This ascription of desire to the building is legitimate, as it helps us to conceptualize the building and “understand [...] its past or future behavior” (McCarthy 1979).

Rabeneck, among others (Fox 2010; Fox and Kemp 2009). While these forward-thinking individuals

developed many concepts between the 1960s and early 1990s, it was not until the mid-1990s that

computational power became affordable enough for physical exploration of those concepts to

be widely available to students and practitioners of architecture. An increase in accessibility and

affordability of microprocessors, such as the Arduino prototyping environment (Arduino 2012), has

generated an explosion of kinetic, responsive, interactive architectural projects in recent years (Fox

1): in response to a certain state of the environment, the system takes an action that it assumes appropriate. Upon completion of the action, it receives a reward representing the goodness of its action, which is then used to update its knowledge of the environment. When presented with a similar situation, the system can now use this updated knowledge to make a better decision toward its goal of maximizing the rewards and thereby finding the optimal control law.

in carrying out required activities” (Chan et al. 2008). Some notable projects from this field include

the TRON Intelligent House (Roppongi, Tokyo), the EasyLiving project (Microsoft), the AwareHome

(Georgia Tech), the House_n project (MIT), and the Neural Network House (University of Colorado).

The work done in the IE sphere has tended to focus more on the networking of devices and

and Negroponte, in order to develop architectural systems with the ability to evolve their behavior

Our interest lies in building upon and bridging the work of the IE community with that of Pask, Price,

To explore our ideas of adaptivity in architecture, we turn to the building envelope. The building

2.2 Scaling Up: Distributed Adaptive Systems

animals are trained, and can be generally referred to as a trial-and-error approach (see Figure

defined as “…any living or working environment that has been carefully constructed to assist people

multimedia applications, and has not focused on architectural design per se.

goals through interaction with its environment. This happens similarly to how humans learn or pet

environments (IE) or smart homes. While the nomenclature varies, these projects can broadly be

Barto 1998). Simply put, an RL algorithm finds the control law that optimally balances different

this paper, and the interested reader is referred to standard textbooks (Mitchell 1997; Sutton and

specifically reinforcement learning (RL) algorithms. A detailed description is beyond the scope of

Parallel to the development of interactive architecture has been the development of intelligent

such as Usman Haque, Ruairi Glynn, and Skylar Tibbits, to name a few.

recently, the work of Gauge and Fox has been built upon by interaction designers and architects

Stephen Gauge at the Bartlett, and the Kinetic Architecture Group led by Michael Fox at MIT. More

To achieve this type of adaptivity, we propose to use algorithms from the field of machine learning,

than the occupant’s actual desires. Ideally, the systems could learn through user interaction and

intelligence and robotics into architecture. This history includes the works of Brody, Eastman, and

and Kemp 2009). Early pioneers of this new wave included the Interactive Design Workshop led by

consequences for building occupants, e.g., using set points based on mean values for comfort rather

people and projects represent only a small fraction of the long and rich history of integrating artificial

improving building energy performance can and should be done in a way that does not have negative

design strategies is the negotiation of building energy performance and occupant desires. In fact,

One contemporary issue in architecture that can be addressed through the integration of adaptive

as opposed to preconceived assumptions, and therefore offers context-appropriate performance.

Negroponte’s work was also based on the ideas of his predecessors, and the previously mentioned

between the gerbils and the robotic gripper evolved the physical environment (Negroponte 1975).

realign them in their new location if they had been significantly displaced. Over time, this conversation

robotic arm would straighten those cubes that had been rotated slightly from their original position, or

curious gerbils roamed about the metal cube environment, they bumped and moved the cubes. The

Adaptivity in architecture allows buildings to change their behavior in response to real-world events

on,” i.e., the threshold light level is changed without explicit human reprogramming, then we speak

Nicholas Negroponte within the Machine Architecture Group at MIT (Negroponte 1970). Specifically,

robotic gripper with a set of small metal cubes and a small population of gerbils. As the inherently

However, if over time this control law evolves to “if the light level drops below 472 lux, turn the light

the light on.” In terms of a control system, this is a simple control law, and there is no adaptation.

execute an action due to a certain sensor input, e.g., “if the light level drops below 500 lux, turn

termed responsive. Responsive systems are often referred to as being adaptive because they

notion of preprogrammed motion, termed kinetic, or systems that move based on sensor readings,

Our understanding of an adaptive system is very specific and goes beyond the typical architectural

2.1 On Adaptivity

ADAPTIVE DISTRIBUTED ARCHITECTURAL SYSTEMS

Generator project was unprecedented, many of the core concepts had been previously developed by

2

4 concludes the paper.

produced during two weeklong bachelor-level design workshops on the topic of ASE design. Section

1:1 scale prototype of the adaptive solar envelope (ASE), and highlights some of the student work

component can be scaled up into a distributed system. Section 3 documents the development of the

approaches from machine learning can be incorporated into architecture, and how an adaptive

paper explores some of the issues involved in developing such systems. Section 2 describes how

and this conversation in turn makes possible the negotiation of potentially conflicting desires. This

over time. This ability opens the possibility for a conversation between architecture and occupant,

While the scale of execution conceived by Price for the self-reconfigurable architecture of the

(Riley et al. 2002).

could be a conversation, so to speak, between occupant and architecture, each influencing the other

configurations on its own. Because of this self-motivated reconfiguration by the architecture, there

be rearranged by its users to suit their desires. In addition, the architecture could suggest novel

of a reconfigurable architecture. The core concept of the project was that the architecture could

Price on his Generator project (1978–80). They helped Price develop digital and physical models

Previous to leading Diploma Unit 11, John and Julia Frazer worked as consultants for Cedric

Unit 11 at the Architectural Association.

architecture, the ideas embedded in Pask’s statement predate the work of John Frazer’s Diploma

practice for many. And while this quote from Pask is often cited as a root of contemporary digital

architecture operates as an intelligent system is still science fiction for some, it has become daily

three decades, inspiring architects to design from a systems perspective. While the idea that an

evolve” (Frazer 1995). This sentence has reverberated through the architecture world for close to

here, I think, is not so much to design a building or city as to catalyse them; to act that they may

As Gordon Pask wrote in the foreword to An Evolutionary Architecture, “The role of the architecture

Agent/environment interaction. Adapted from Sutton and Barto 1998.

figure 1

figure 1

// Synthetic Optimizations

449

450

Scales of distribution: a) single component scale; b) cluster scale; c) window scale; and d) building scale. Varying degrees of autonomy and control can be applied at various scales.

figure 3

Various states of building envelope components: a) closed state, reflecting solar radiation; b) open state, allowing views and bouncing daylight into the interior; c) solar tracking state, optimizing for maximum power generation if PV elements are integrated; and d) “mixed state,” allowing multiple functions to be fulfilled simultaneously. It is the flexibility of the mixed state that gives a system the ability to adapt to user desires.

figure 2

ACADIA 2012 // Synthetic Digital Ecologies

3

figure 4

figure 3

figure 2

The prototype is 2.25 m high by 1.25 m wide and consists of 36 modular components. The movement

energy, kinetics, user interaction, adaptive control, and architectural design.

As shown in Figure 4b, this project is highly interdisciplinary, integrating concepts of renewable

multiple aspects related to the implementation of adaptive distributed systems in built architecture.

The Adaptive Solar Envelope (ASE) 1:1 scale prototype, shown in Figure 4a, was developed to explore

3.1 Physical Prototype

ADAPTIVE DISTRIBUTED DESIGN

envelope, which expresses the negotiation between building and occupant desires.

negotiating conflicting goals. The result of such a system also produces a visually dynamic building

variation in response to internal and external factors, and thereby offers the highest probability of

locations depending on the goal of the system. We believe that a distributed approach allows for local

envelope of a building (see Figure 3). This distributed approach can be deployed at various scales and

window or particular interior spaces. These, in turn, if proliferated, could make up the facade or entire

it can also be clustered into small groups. Likewise, groups of clustered modules can relate to a

the wiring between the sensors and the motors the vehicles would express a “light loving” behavior

pair of light sensors to drive a pair of motors. As shown in Figure 7, depending on the relationship of

be observed in manmade artifacts. One of the vehicles developed conceptually in the book used a

Braitenberg described how through building simple sensor/actuator relationships “behaviors” could

“Gedankenexperiment” Vehicles: Experiments in Synthetic Psychology (Braitenberg 1984).

The initial inspiration for our solar tracking module came from Valentino Braitenberg’s

and Nijmeh 2004).

year, which can increase power production from photovoltaic systems by 35–40 percent (Abdallah

gives the components the flexibility to track solar movements throughout the course of the day/

6c shows the hemispherical operational boundary allowed by the setup. This freedom of rotation

brackets, giving the module two degrees of rotational freedom, shown in Figures 6a and 6b. Figure

of the basic module are shown in Figures 5a–c. The stacked servo pair is mounted with pan-tilt

servos paired with light sensors, a microcontroller, and a solar thin film cell. The various elements

a preprogrammed series of movements. The basic module of the ASE consists of a pair of stacked

of the components can be defined by user control or by a solar tracking behavior, or it can run in

figure 6

figure 5

a) 180° rotational axis of lower servo; b) 180° rotational axis of upper servo; c) Combined rotational axes of servos creating a hemispherical operational boundary.

figure 6

a) First prototype of solar tracking component combining light sensors with a pair of stacked servos and a microcontroller; b) stacked servo pair with custom solar thin film mounting bracket; and c) solar thin film connected to a trickle charging battery circuit.

act independently to optimize shading or solar harvesting potential. In addition to acting autonomously,

figure 5

If we imagine that a facade element has a certain behavior—for example, solar tracking—then it can

a) ASE 1:1 prototype; b) Interdisciplinary nature of the ASE project.

figure 4

cooling loads, and improve distribution of daylight.

in Figure 2a–c, the envelope can mitigate solar insolation, thereby offering reductions in heating/

area in terms of architectural design, as it is, in essence, the public face of a building. As shown

through which an architecture and an occupant may converse. The envelope is also a highly important

envelope, acting as a buffer between interior and exterior environments, offers an ideal medium

// Synthetic Optimizations

451

452

figure 7

a) Semiregular tiling pattern chosen for ASE prototype; b) Structural geometry defined by center points of tiling pattern; c) Cluster of modules with support structure; d) Geometric pattern of the support structure; e) Proliferation of modules.

figure 9

Regular tiling geometries a, b, and c. Semiregular tiling geometries d, e, and f.

figure 8

Braitenberg vehicles. The vehicle on the left has crossed connections between its sensors and motors and exhibits a “light loving” behavior. The vehicle on the right has straight connections between its sensors and motors and exhibits a “light phobia” behavior. Adapted from Braitenberg 1984.

figure 7

ACADIA 2012 // Synthetic Digital Ecologies

figure 9

figure 8

tiling patterns, shown in Figures 8d–f, offer increased visual complexity without the need for entirely

and hexagons are the three basic nesting tiling geometries, shown in Figures 8a–c, but semiregular

degrees of translucency, which widens the design palette open to architects. Triangles, squares,

manufacturers offer the production of custom geometries and an increasing number of colors and

The tiling pattern of the modules is not confined to a predetermined size or geometry. Solar thin film

to site-specific and time-specific events.

responds to local instances of overshadowing. This localization of behavior is essential in adapting

makes its installation location independent of programming, and also means that its behavior

preprogramming solar tracking paths based on geographic location. The behavior of the module

and Schlueter 2012). This setup gives the component the ability to track solar movements without

control algorithm to teach the module that its goal is to follow the movement of the sun (Rossi, Nagy,

top of the first in a pan-tilt arrangement (see Figure 5b). We implemented a reinforcement learning

Once a single Braitenberg servo was operating, a second servo with the same setup was stacked on

The rotational direction of the servo is defined by comparing the readings from two light sensors.

We adapted the basic setup for a Braitenberg vehicle by replacing the two motors with a servo motor.

or a “light phobia” behavior.

The finished prototype ran as an installation for one month. At various times it operated in solar-

custom mounting solution that was easy to assemble and install (see Figures 11 and 12).

rapid back-and-forth between design and physical prototype allowed for the quick development of a

out through the use of CNC fabrication techniques including laser cutting, milling, and 3D printing. The

The production of scale models and the iterative exploration of module mounting solutions were carried

produce many variations, as shown in Figure 10.

rather than through nonstandard components. Even a regular tiling pattern has the potential to

and fabricate complex geometries. The complexity of the system is generated through its behavior

external environmental influences, complex surface patterns can emerge without the need to design

Because the components move independently of each other, and in response to varying internal and

of each module defined a connection location between it and the support structure

tiling pattern was also responsible for defining the underlying support structure, as the center point

a high degree of visual interest through the use of only two repeating geometries. The semiregular

regular tiling pattern, shown in Figure 9, to develop into the built prototype. The pattern produces

In order to explore the aesthetic potential of custom solar thin film geometries, we selected a semi-

nonstandard production of components.

figure 11

figure 10

a) Component assembly of ASE 1:1 prototype; b) Mounted assembly.

figure 12

Potential states of adaptive envelope modules with control expressed at different scales: a) building scale, closed; b) building scale, open; c) building scale, solar tracking; d) window scale, mixed; e) cluster scale, mixed; f) individual module scale, mixed.

figure 11

a) Semiregular tiling pattern chosen for ASE prototype; b) Structural geometry defined by center points of tiling pattern; c) Cluster of modules with support structure; d) Geometric pattern of the support structure; e) Proliferation of modules.

figure 10

figure 12

// Synthetic Optimizations

453

454

a) Physical prototype of adaptive facade component; b) Interior renderings of adaptive facade system in various states. Project by Eva Lüginbuhl and Fabian Reimer.

figure 14

ASE 1:1 prototype in various states: a) closed state; b) open state; c) solartracking state; d) three-cluster mixed state; e) six-cluster mixed state; and f) autonomous module mixed state.

figure 13

ACADIA 2012 // Synthetic Digital Ecologies

4

how distribution allows for both architectural expression and negotiation of potentially conflicting

insolation. Figure 14b shows interior views with the facade system in various states.

goals of occupant desires and building energy performance. In two weeklong workshops we tested

In this paper, we investigated the design of an adaptive distributed building envelope. We showed

SUMMARY AND CONCLUSIONS

education is key to training architects in the design of adaptive distributed architectural systems.

Shown in Figure 14a is the physical prototype of a component that could spiral open to reduce solar

into facade systems with the ability to track changing solar angles and adapt to occupant desires.

of two developed adaptive solar facade components. The components were conceptually developed

components. Students were introduced to the Arduino prototyping platform, and worked in groups

architectural systems. We believe that introducing these concepts and techniques early in architectural

unfamiliar modeling and physical prototyping platforms to apply them to developing adaptive distributed

The first seminar week focused on introducing physical computing as a means to develop adaptive

In spite of the short timespan of these workshops, students were able to become sufficiently fluent in

in the concepts of adaptive architectural design.

degrees in response to occupant desires.

during the workshop. The system can be fully closed for optimal solar harvesting, or opened to varying

in Figure 15 show interior and exterior representations of an adaptive facade system developed

developed into facade systems with the ability to adapt to internal and external influences. The images

students developed parametrically controlled components. These components were conceptually

Each workshop took a slightly different approach to engaging undergraduate architecture students

students to adopt the concepts and techniques involved in developing adaptive architectural systems.

adaptive solar facade design. The workshops were conceived to determine the ability of architecture

In addition to developing the ASE prototype, we have led two weeklong workshops on the topic of

3.2 Teaching Adaptive Distributed Design

modeling to more quickly develop adaptive facade concepts. Through the use of Rhino/Grasshopper,

its various possible configurations, shown in Figure 13 (Adaptive Systems Lab 2012).

figure 15

During the second workshop we engaged a different group of students in the use of parametric

tracking mode or user-controlled mode, or ran a preprogrammed series of movements to demonstrate

figure 14

figure 13

a) Interior rendering of parametrically controlled adaptive solar facade system; b) Exterior rendering of adaptive solar facade system. Project by Martin Wey and Thorben Weisterhuys.

figure 15

// Synthetic Optimizations

455

456

ACADIA 2012 // Synthetic Digital Ecologies

Sutton, R. S., and A. G. Barto. (1998). Reinforcement Learning: An Introduction. Cambridge, MA: MIT Press.

Rossi, D., Z. Nagy, and A. Schlueter (2012, in press). Adaptive Distributed Robotics for Environmental Performance, Occupant Comfort and Architectural Expression. International Journal of Architectural Computing (IJAC).

Riley, T., et al. (2002). The Changing of the Avant-Garde: Visionary Architectural Drawings from the Howard Gilman Collection. New York: Museum of Modern Art.

Negroponte, N. (1975). Soft Architecture Machines. Cambridge, MA: MIT Press.

Negroponte, N. (1970). The Architecture Machine. Cambridge, MA: MIT Press.

Mitchell, T. M. (1997). Machine Learning. New York: McGraw-Hill.

McCarthy, J. (1979). Ascribing Mental Qualities to Machines. Accessed June 27, 2012. http://www-formal. stanford.edu/jmc/ascribing/ascribing.html.

Frazer, J. (1995). An Evolutionary Architecture. London: Architectural Association.

Fox, M. (2010). Catching Up with the Past: A Small Contribution to a Long History of Interactive Environments. In FOOTPRINT: Digitally-Driven Architecture, eds. H. Bier and T. Knight, spring 2010, 5–17.

Fox, M., and M. Kemp. (2009). Interactive Architecture. New York City: Princeton Architectural Press.

Chan, M., D. Estève, C. Escriba, and E. Campo. (2008). “A Review of Smart Homes—Present State and Future Challenges.“ Computer Methods and Programs in Biomedicine 91: 55 –81.

Braitenberg, V. (1984). Vehicles: Experiments in Synthetic Psychology. Cambridge, MA: MIT Press.

Arduino. (2012). Accessed June 27, 2012. http://www.arduino.cc.

into the fabric of the built environment at early design stages when decisions are most influential.

is facilitated. Synergies of embodied carbon, material, form, wind, seismicity, etc. can be incorporated

When these tools are applied to multiple buildings at a district scale, a new level of design and planning

building systems such as structure, mechanical, elevatoring, etc.

(PCM). PCM estimates the usable area of a tower by computing net floor area considering major

carbon dioxide emissions of structural components. The second is Parametric City Modeling

The first is the Environmental Analysis Tool™ (EA Tool). EA Tool quantifies the estimated equivalent

stories, occupancy, material type and quantities, geographic location, seismicity, climatic influences, etc. Utilizing collected data, two informative analysis tools for intelligent design are proposed.

Adaptive Systems Lab (ASL). (2012). Accessed June 27, 2012. http://adaptivesystemslab.blogspot.ch/.

could inform current design practices. These parameters include gross floor area, number of

Designers have examined attributes of previously designed projects to understand how key parameters

synergy-oriented environments.

resource of design knowledge which, when brought into fruition, can guide designers toward efficient,

necessitates a new methodology of design. Data describing constructed buildings is an untapped

An ever-changing state of design provoked by unpredictable economies and shifting cultures

ABSTRACT

PE, LEED AP Skidmore, Owings & Merrill LLP

David Shook

PE, SE, LEED AP, Director Skidmore, Owings & Merrill LLP

Mark Sarkisian

SYNTHESIZING ELEMENTS FOR TALL BUILDING DESIGN

WEIGHTED METRICS:

Abdallah, S., and S. Nijmeh. (2004). Two Axes Sun Tracking System with PLC Control. Energy Conservation & Management 45: 1931–39.

REFERENCES

contributions to the development of the ASE 1:1 prototype.

We would like to acknowledge our student assistants Eva Lüginbuhl and Julien Bellot for their

ACKNOWLEDGMENTS

wrote, and before long, adaptive architectures will be commonplace in our built environment.

are exposed to these concepts and techniques, they are bound to become the catalysts of which Pask

interdisciplinary nature. However, as an increasing number of architecture students and practitioners

architecture that is largely unexplored. The reason for this may be the steep learning curve of its

environmental performance, user satisfaction, and aesthetic possibility, it remains an area of

While the integration of adaptive systems into architecture holds great potential in relation to

parametric modeling platforms.

to implement those concepts into design projects through the use of physical computing and

the capacity of bachelor-level architecture students to grasp the concepts of adaptive design and

// Synthetic Optimizations

457