The Implementation of Programmable Architecture: Wireless Interaction with Dynamic Structure


-Wireless interaction with dynamic structure



K. HOTTA 1 and A. HOTTA 2
1,Keio University, Fujisawa, Japan 1 University of Tokyo, Tokyo, Japan;
2,Kagawa university, Kagawa, Japan

1,堀田憲祐 2,堀田明登


True adaptability in architecture necessitates both dynamic hardware and software with the potential for continually renewable forms capable of all possible variations necessary for changing demands and conditions, without having to resort to one theoretically optimal solution. PA consists of both autonomous and subservient systems that maintain a constant homeostasis within its contained environment. The information flow between the Genetic Algorithms (GA) and user input prompts this hybrid system to generate the consequent, ever-changing physical form, while continuously optimizing it for environmental stimuli. This paper proposes a smart strategy for a human interactive-cybernetic architecture in the context of K. Hotta’s Programmable Architecture (PA), aimed at enhancing GA’s capabilities in continuous self-modelling and facilitating human-computer interface.




#Human-computer interaction, #user interface.


#ヒューマンコンピュータインタラクション #ユーザーインターフェース

1. Introduction

In recent years, the key word ‘adaptability’ has increasingly been used in the field of emergent architectural design such as in the late book titled ‘Adaptive Ecologies’ (Theodore et al, 2013). Also, one of the most established architectural conferences, ACADIA (The Association for Computer Aided Design in Architecture), titled its 2013 symposium ‘Adaptive Architecture’. There are many approaches to making architecture (both buildings and sys- tems) adaptive and sustainable. Most contemporary architecture uses scientific approaches based on mathematics, physics, and computational tools and through experimentation is able to achieve higher levels of adaptability against ever changing circumstances. However, in considering the keyword ‘adaptability’ in architectural design, it is critical to discuss what adaptability is for. In the previous paper of PA (Hotta, 2013), environmental adaptability is the centre of discussion. The aim is to develop architecture with higher environmental adaptability, and to develop an architectural design methodology that will address issues of shape, systems, and devices that constitute the building.

1. 序論

近年、「Adaptive Ecologies」(適応生態学 )(Theodore et al, 2013)という書籍が出版されるなど、創発的建築設計の分野で「適応性」というキーワードが使われることが多くなってきている。また、最も定評のある建築学会の一つであるACADIA(The Association for Computer Aided Design in Architecture)は、2013年のシンポジウムを「Adaptive Architecture」(適応建築学 )と題して開催している。建築(建物とシステムの両方)を適応的かつ持続可能なものにするためには、多くのアプローチがある。現代の建築の多くは、数学、物理学、計算ツールに基づく科学的アプローチを用い、実験を通して、絶えず変化する状況に対してより高いレベルの適応性を実現することが可能である。しかし、建築設計において「適応性」というキーワードを考える上で、何のための適応性なのかを議論することは非常に重要である。PA の前論文(堀田, 2013)では、環境適応性が議論の中心となっている。より環境適応性の高い建築を目指し、建築を構成する形状、システム、デバイスの問題を解決するための建築設計手法を開発することを目的としている。

2. State of the art

In the architecture field, there is a seminal book in terms of participatory planning methods; an architect-less architectural system in the vernacular tradition (Rudofsky, 1964). More recent, with reference to temporal design linked with the idea of emergence, it is worth looking at the Fun Palace Plan (Price, 1961) and Plug-in City (Archigram,1964) in the United Kingdom, and the Metabolism Movement in Japan in the 1960s and 70s (Lin, 2010). The idea of emergence in the previously mentioned contexts is that the architect and designer designed ‘systems’ rather than depicting static images. On the other hand, there are many precedents in the engineering field. For example, Cybernetics, advocated by Norbert Wiener (1961), was a synthetic academic discipline that dealt with the matter of control and correspondence in a system like an organism or a machine. As an extension of this line of thought, Control Theories describe the methods in engineering and mathematics, which aim to control dynamic behaviour. The usual objective of control theory is to control a system. It attempts to adjust the system behaviour through the use of feedback via a controller, further developed as P, PI, PID (Proportional, Integral, Derivative) (Minorsky, 1922). In the robotics field, the ‘smart’ robotic system has its origins in Social Robot (Walter, 1951), where the first speculation of multiple interacting robots was conceived. Another early trial of ‘smart’ robot can be referred to in ‘Vehicles’ (Braitenberg, 1984), in which evolutionary thinking was implemented in the wiring itself. Subsumption Architecture (SA) is the reactive idea used in artificial intelligence (AI), a term and field invented by R. Brooks (1986), which was developed to determine robot behavior. Swarm robotics is a relatively recent methodology for a multiple robot system (Yim et al, 2000), based on swarm intelligence observed in animal behavioral patterns. These past and seminal works ultimately have led to Evolutional Computing (Sims, 1994) and Evolutional Robots (Lipson, 2000).


建築の分野では、参加型計画の手法という意味で、ヴァナキュラーの伝統に基づく建築家不在の建築システム(Rudofsky, 1964)という画期的な本がある。最近では、創発の思想と結びついた時間的デザインとして、イギリスの「ファンパレス計画」(プライス、1961)や「プラグインシティ」(アーキグラム、1964)、日本の1960〜70年代の「メタボリズム運動」(林、2010)などが参考になる。先に挙げた文脈における創発の考え方は、建築家やデザイナーが静的なイメージを描くのではなく、「システム」をデザインしたことである。一方、工学の分野でも多くの先例がある。
例えば、Norbert Wiener(1961)が提唱したCyberneticsは、生物や機械のようなシステムにおける制御と対応の問題を扱う総合的な学問分野であった。その延長線上にあるのが「制御理論」であり、動的な振る舞いを制御することを目的とした工学や数学の手法である。制御理論の通常の目的は、システムを制御することである。P、PI、PID(Proportional, Integral, Derivative)(PID制御)としてさらに発展したコントローラを介してフィードバックを利用し、システムの挙動を調整しようとするものである(Minorsky, 1922)。ロボット工学の分野では、「スマート」なロボットシステムはSocial Robot (Walter, 1951) に起源を持ち、そこで初めて複数の相互作用するロボットの構想が練られた。また、「スマート」ロボットの初期の試みは、進化的思考を配線自体に実装した「ビークル」(Braitenberg, 1984)に言及することができる。サブサンプション・アーキテクチャ(SA)は、R. Brooks(1986)が考案した
人工知能 (AI)で用いられる反応的な考え方で、ロボットの行動を決定するために開発された用語・分野である。スワームロボティクスは、動物の行動パターンに観察されるスワームインテリジェンス(群知能)に基づく、比較的最近の複数ロボットシステムの方法論である(Yim et al, 2000)。これらの過去の代表的な研究は、最終的にEvolutional Computing (進化的計算)(Sims, 1994) や Evolutional Robots(進化ロボティクス) (Lipson, 2000) につながっている。

3. Background

As a case study of the PA, flexible kinetic-tensegritic structures are shown in previous thesis (Hotta, 2013). The hardware for this proposal is an accumulation of self-sufficient machines that is dedicated to the actions of sensing, calculating, and actuating. As a case study for this thesis, a kinetic canopy that is organized using tensegrity-based components of variable forms is proposed. This architectural robot is actuated by shape memory alloy (NiTi) instead of tensile wire, and its control is handled electrically by micro controllers Arduino (Banzi et al, 2005~). A physical model of this machine has been built at a one-to-one scale and user-tested via mobile devices such as a smart phone.

3. 背景

PA (プログラマブルアーキテクチャ)のケーススタディーとして、フレキシブルなテンセグリティ構造が以前の論文で示されている(堀田, 2013)。本提案のハードウェアは、感知・計算・作動という動作に特化した自給自足的な機械の集積である。本論文のケーススタディとして、テンセグリティに基づく可変形状の部品を用いて編成されたキネティックキャノピーを提案する。この建築ロボットは、テンシルワイヤーの代わりに形状記憶合金(NiTi)を用いて作動し、その制御はマイクロコントローラArduino(Banzi et al, 2005〜)により電気的に処理される。この機械の物理モデルを1対1のスケールで製作し、スマートフォンなどのモバイルデバイスを用いたユーザーテストを行った。

Figure 1  画像

Figure 1. 4th Generation, Physical Model (referring from PA, Hotta, 2013).

図1. 第4世代、物理モデル(PA, 堀田, 2013を参照)

The software for this proposal consists of a hybrid control system, which attempts to minimize the difference between the desired objective values and the measured values. This is a combination of automatic responses and user manipulations in order to achieve a faster and higher degree of adaptation. Utilizing the versatility of GA, multiple user inputs are proposed to partially substitute for its purely random mutations (usually GA use random digits for mutations).



Figure 2. Drawing of PA (referring from PA, Hotta, 2013).

図2. PA の図面(PA, 堀田, 2013 を参考にした)。

This resolves GA’s shortcomings, namely protracted calculation time, lack of adaptability to a fluctuating objective function which represents the ideal condition at any given time, and the ability for ad hoc responses when the system experiences usage overload or random environmental fluctuations. Incorporating the user input, the system can respond rationally to actual conditions unanticipated by the GA. Therefore, the user can concurrently control the system locally, to reflect individual preferences, and contribute to the global optimization and increased efficiency of the system as a whole.



Figure 3. Experiment-1 and its result (referring from PA, Hotta, 2013).

図3. 実験-1とその結果(PA, 堀田, 2013より引用)。

The exposure experiment-1 executes with 4 candidates, which consists of Fixed, Pre-optimised (with same algorithm), Realtime-Kinetic (Proposed PA system), Kinetic ultimate (ideal score). The aim of these experiments was to prove that the proposed kinetic roof adapts most effectively to sun exposure over time providing the best result. This set of GA experiments demonstrate the result that normal GA does not work effectively for dynamic situations. The score for the real-time kinetic roof was worse than the pre-optimized roof’s score. The following paragraph discusses the possible reason for this and suggests potential solutions. Usually GA is designed and used for static problems. The key issue to focus on in order to improve the score is how to compensate for the lack of initial information in a moving landscape. Two possibilities are: Insert sensor information into the GA’s ongoing calculation as an interruption input.



Figure 4. GA+Human Input Concept Diagram (referring from PA, Hotta, 2013).

図4. GA(遺伝的アルゴリズム)+人間による入力の概念図(PA, Hotta, 2013より引用)。

On those contexts, Experiment-2 has examined. While the previous model only allowed one input, namely the environmental input from the sun, this model has 2 inputs, one the top-down environmental input similar to the GA, and the other being a bottom-up user input explained below. Obviously, this is not a pure GA system, but rather a human-interactive evolutionary algorithm. Instead of ‘random’ mutation, user input is used to seed the next generation. This will help reduce the calculation time.


Figure 5. PA Simulation Result (referring from PA, Hotta, 2013).

図 5. PAシミュレーション結果(PA, 堀田, 2013より引用)。

Five different executions are shown as graphs; bigger number is better. Each graph shows a unique ‘GA+Human’ method, the three candidates: (Fixed, Real-time-Kinetic, Kinetic-Ultimate) are the same. Each ‘GA+Human’ method’s score depends on the proficiency of the human and the timing (number of interventions.). The conclusion of this experiment is ‘GA+Human’ not always result better than normal GA (Realtime Kinetic) but, sometime could beyond it. This shows if the human manipulations were appropriate, this system can exceed normal GA system. Discrepancy points between the first model and this model are there because those are made by different settings but this is not a main point of this paper. The Grasshopper experiment raised issues and then it made hypotheses that have been tested in this chapter. In this chapter, the mathematical model (Processing code) confirmed the same minimum set of points; showing that normal GA does not work effectively where dynamic fitness calculations are required. Here a model using Interactive Evolutionary Computation (IEC) was designed and tested. While reaching similar conclusions these chapters could be said to be logically independent.

5種類の実行結果がグラフで表示され、数字が大きいほど良い結果である。各グラフは、GA+人間による独自の手法であり、3つの候補(固定、リアルタイムキネティック、最終キネティック)は同じものである。それぞれの「GA+人間」手法のスコアは、人間の習熟度とタイミング(介入回数)によって変化する。 この実験の結論は、「GA+人間」が常に通常のGA(リアルタイムキネティック)よりも良い結果を出すとは限らないが、時にはそれを超えることもあり得るということだ。これは、人間の操作が適切であれば、このシステムは通常のGAシステムを超えることができることを示している。なお、最初のモデルと本モデルでは設定が異なるため、矛盾する点があるが、これは本論文の主旨ではない。グラスホッパーの実験が問題提起となり、本章で検証する仮説が生まれた。本章では、数学的モデル(Processing code)により、同じ最小限の点を確認し、動的な適性計算が必要な場合には、通常のGAが有効に機能しないことを示した。ここでは対話型進化計算(IEC)を用いたモデルを設計しテストした。同じような結論に達しているが、これらの章は論理的に独立していると言える。

4. Problem statement

Reviewing previous PA research (K.Hotta, 2013), several shortcomings raise up. Here is the problem statement. Human-machine hybridized systems yield relatively better results in the simulation process. However, the large quantity of sliders and lack of intuition in the interface hinders usability and efficiency.

4. 問題提起

これまでの PA 研究(K.Hotta, 2013)を見直すと、いくつかの欠点が浮かび上がってくる。以下は問題提起である。人間と機械をハイブリッド化したシステムは、シミュレーションプロセスにおいて比較的良い結果をもたらす。しかし、大量のスライダーと直感的なインターフェイスの欠如が、使い易さと効率性を妨げている。

5. Physical implementations

In order to get higher score, several systems are tested below.



Figure 6. PA model version-1 and its system.

図6. PAモデルバージョン-1とそのシステム

The early model of PA, which was made by Arduino in ‘standalone mode’, used the PC strictly as a power source. Though receptive and relatively quick, it was limited to simple responsive or at most branching systems written by the syntax ‘if’, ‘for’, ‘case’ etc. The crucial shortcoming is there is no human interface.



Figure 7. PA model and its system version 2.

図7. PAモデルおよびそのシステムバージョン2

The wireless system was tested with TouchOSC (, 2012), which was originally designed for MIDI controlled surfaces. Yet, its limitations were revealed in its inability to combine two sets of data.

ワイヤレスシステムのテストには、もともとMIDI制御のサーフェス用に設計されたTouchOSC (, 2012) が使用された。しかし、2つのデータを組み合わせることができないなど、その限界は明らかだった。

Figure 8. PA model and its system version 3.

図 8. PA モデルとそのシステムバージョン3。

This implementation is the minimum set for PA, which can afford hybrid data flow, perform automatic calculations and respond to human input. However, the wireless interface with an end device, such as a smartphone, could not be created in reality.



Figure 9. PA model version 4.

図9. PAモデルバージョン4。

Referring to Bongard’s idea of continuous self-modelling (Bongard, 2006), PA was utilized for simulating precise and physics based virtual models (Fig.10), which was used for previous PA experiments. Ultimately, this will result in the resistance against environmental catastrophic change such as earthquakes, typhoons, etc... Additionally, by selecting outputs instead of expressing every genome could be expected to reduce both energy and cost.

Bongard(ボンガード)の連続的な自己モデリングという考え方(Bongard, 2006)を参考に、これまでのPA実験で使用されてきた精密で物理に基づいた仮想モデル(図10)をシミュレーションするためにPAを活用した。最終的には、地震や台風などの環境破壊的な変化に対する抵抗力を高めることになる。また、すべてのゲノムを表現するのではなく、出力を選択することで、エネルギーとコストの削減が期待できる。

6. Experiment and result

Fig. 11 illustrates the comparison between the two different types of user interfaces. The ‘simple sliders’ (pictured left, Fig. 10) which control each genome independently, yield less performative results, as oppose to the ‘smart panel button’ (pictured right, Fig. 10), which connects 6 clouds of actuators into a single component, thus demonstrating that the panel control is more intuitive and corresponds accurately with the virtual modelling space.

6. 実験と結果



Figure 10. Two types of Interfaces; Left is sliders, Right is panels.

図10. 左がスライダー、右がパネルの2種類のインターフェイス


Figure 11. The result of experiment, bigger is better.

図11. 大きい方が良いという実験結果

7. Conclusion

Though advances in computational simulations have continuously advanced in recent years, their integration in human-computer interface and dynamic adaptability often lend architectural models to remain as static outputs. The smarter UI strategy is proposed as a step towards an adaptable architecture in two ways. First, by facilitating the computer- human interface in an intuitive modeling process, and second, by favoring a continuous dataflow between the virtual model and physical model, rather than a simple GA setup.

7. 結論

近年、計算シミュレーションの進歩は留まるところを知らないが、ヒューマンコンピュータインタフェースとの融合や動的適応性により、アーキテクチャモデルが静的なアウトプットに留まってしまうことが少なくない。本論文では、2 つの観点から、よりスマートな UI 戦略を提案する。第一に、直感的なモデリングプロセスにおいてコンピュータと人間のインターフェースを容易にすること、第二に、単純なGA設定ではなく、仮想モデルと物理モデルとの間の連続的なデータフローを支持することである。


I would like to express my thank to Mr. M. Weinstock and Dr. G. Jeronimidis, as an academic supervisors, I. Wako and Y. Fujimaki as drawer. Also great thank for Union foundation as a sponsor.


指導教官であるM. Weinstock氏、G. Jeronimidis氏、図面を引いてくれたI. Wako氏、Y. Fujimaki氏に感謝の意を表す。また、スポンサーであるユニオン財団に感謝する。

References 参考文献

Banzi, M. and Cuartielles, D., IGOE and T., Martino, G. and Mellis, D.: 2005~, “Arduino”. Available from: (accessed 2013).

Bongard. J and Zykov.V and Lipson.H ,:2006, Resilient Machines Through Continuous selfmodelling, Science Magazine, 17Nov, 1118–1121. Braitenberg, V.: 1984, Vehicles. The MIT Press, Cambridge, USA.

Brooks, R.: 1991, New Approaches to Robotics, Science, 253, 1227–1232. Cook, P.: 1999, Archigram, Princeton Architectural Press, USA.

Hotta, K.: 2013, Programmable Architecture, AA School, UK.

Holland, J.H,:1975, Adaptation in Natural and Artificial System, University of Michigan.

Huber, J.E., Fleck, M.F. and Ashby: 1997, The selection of mechanical actuators based on performance indices, The Royal Society, UK.

Lin, Z.: 2010, Kenzo Tange and Metabolist Movement, Routledge, London UK.

Lipson, H and Pollack, J.: 2000, Automatic Design and Manufacture of Artificial Lifeforms., Nature, 974-978.

Minorsky, N.: 1922, Directional stability of automatically steered bodies, J. Amer. Soc. Naval Eng.

Mtchell, M.: 1996, An Introduction to Genetic Algorithm, MIT Press, Cambridge.

Price, C.: 2003, Cedric Price: The Square Book, Wiley-Academy, London.

Rudofsky, B.: 1964, Architecture without Architects, Univ of New Mexico Press, Mexico.

Sims, K.: 1994, Evolving Virtual Creatures, Siggraph ’94, Orlando, USA.

Theodore, S. and Frazer, J. and Schumacher, P.: 2013, Adaptive Ecologies: Correlated Systems of Living, Architectural Association Publications, London.

Walter, G.W.: 1953, The Living Brain, Duckworth, London

Wiener, N.: 1961, Cybernetics, The MIT Press, Cambridge, USA.

Yim, M. et al,: 2000, PolyBot: a Modular Reconfigurable Robot, Xerox Palo Alto research Centre, International conference on Robotics & Automation, San Francisco, USA.