EMBEDDED PATTERN An Adaptive SELF LEARNING WORKSPACE Spaces of the 21st century Studio Master Prof. Krassimir KRastev Second Advisor M.arch Alexander kalachev DESSAU INTERNATIONAL ARCHITECTURE Graduate School
I would like first to express my gratitude for my parents and my sister. I would like to thank Prof. Krassimir Krastev for his advices and guidance through my thesis, it has been an honor being in his studio, and as main advisor for my thesis. As well I am very grateful to for M.arch Alexander Kalachev , my second advisor, for all his support through my whole masters. Many thanks as well for all my friends and colleagues in the studio, for their support and help.
II
Contents Studio Brief
1
ABSTRACT 2 INTRODUCTION
3
How AI is affecting the world. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 NEURAL NETWORKS
5
Introduction:. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Perceptron Logic. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 Pattern Recognition Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 Network Logic. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 Adaptive Environment References
19
Lyrical Theatre in Cagliari. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 Shape-shifting sofa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Breathing Wall. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 Auora. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 Bloom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 Thesis Statement
29
Proposal. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 DESIGN PROPOSAL
31
Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 Materials Proposed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 Learning Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Self Autonomous Phase. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 Emerged Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 III
Conclusion
64
Proposed Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 Further Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 Effect on Life style. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 Artificial Intelligence Influence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 References
IV
67
01
Studio Brief Spaces of the 21st Century
The Material Performance Studio is dedicated to exploring the convergence of contemporary architecture with science and technology through computation. Developing deeper practical and theoretical understanding of coding, artificial intelligence and robotics, the studio investigates the emerging transition from culture dominated by mass production towards the postindustrial landscape defined by digital production. This year’s thesis studio will design workplaces for the new industries born by the 21st century’s advances in science and technology, such as tech and software offices or campuses, nano technology labs and research centres; facilities dedicated to biotechnology, renewable energy production, robotics, space exploration etc. Keeping away from “corporate branding”, the studio will explore the impact of innovation in material science and technology onto the way we design, build, and use spaces for work, research and manufacturing. In a mobile and constantly connected working environment, with artificial intelligence and robotics involved in the industry, the way we use our work spaces will change. The technologies that determine the program of the projects will drive the choice of materials, structure, and performance of architectonic elements or assemblages. Understanding innovations in infrastructure, the need for mobility and flexibility; blurring the boundaries between natural and artificial habitats, protecting natural landscapes while praising the liveliness of a community in a neighborhood, the studio will conceive architecture that is informed by technological innovation, as well as by human scale, revealing the materiality of humanity’s integration with technology and science. The studio will investigate the cultural, economic and social conditions and tendencies in the cities established as startup incubators of the tech industry. The students will eventually choose a site for their projects situated in one of the emerging European startup centers, such as, but not limited to: Berlin, Barcelona, Budapest, Amsterdam, Lisbon, Stockholm, Paris, Tallinn, Istanbul, Moscow or Tel Aviv. The studio will travel to one of those cities, the choice to be made upon vote by the students, and study the unique circumstances that can influence the architecture of emerging tech industries. Parallel with that, the studio will research three computational paradigms: Coding, Artificial Intelligence and Robotics. Digital and material experiments on those topics will be recorded and structural prototypes, informed by those studies, will be exhibited at the end of the first semester. Eventually, the findings of the experimental phase will be applied onto a location in a European startup center, developing architecture for the work spaces in the 21st century’s industry. Professor : Krassimir Krastev 1
02 ABSTRACT A brief overview of the thesis
This thesis will be an attempt on how to create fully adaptable workspace, linking architectural design with self-aware (AI) system that its main function is to filter data collected from users, to use it in design decisions. The application will be composed of an interior skin that can perform all the dynamic functionality any workspace can incubates. As like any system it will contains an input and resulting output, which will be:
System Input: The system input will be in the form data collected from user's behaviors in terms of social interaction and their physical change they do inside the space.
System Output: The output will be a behavior reflected on the interior level workspace that can be very dynamic. The adaptation is not only towards a single ideal "optimum", but towards a system in which both the building and the users continuously explore different ways to interact. Using today’s technology, can increase the engagement between people and architecture, shaping the identity of architecture to be a better reflection of the users and their activity inside the workspace. Such radical way of designing will open up an arguments that: • Redefine the way people occupy, or rather interact with buildings. • A speculation of how to offload cognition to building environment.
2
02 INTRODUCTION
Artificial intelligence is being used more and more in many industries can be introduced to architecture
How AI is affecting the world Industries in the 21st century are putting more investment in artificial intelligence technologies, company like Google just bought a startup UK based company called Deep mind that develops mainly artificial intelligence algorithms to be used in gaming for ÂŁ 400 million. Said Demis Hassabis one of Deep
Deep Mind Algorithm can play Attary Breakout
Mind founders that “Climate, disease...AI-assisted science will help the discovery process�. The exponential speed of change happening in the world is affecting all industries making them developing artificial intelligence algorithms to cope with such change . Can such effect also reach the architecture level , structures will be also affected by such rapid speed of change that might require more than just to think of adaptable predesigned structures. Structures in the modern world can be described as temporary buildings that cannot be treated as long lasting with little or no flexibility to it to be modified or even to be changed completely. Making changes in static structures are expensive and efficient. When there will be a renovation, even if some 3
materials might be recyclable, there will be still material waste, leading to a non-sustainable design model. The worst case is that such structures could be even just abounded and just go for a new one, which even waste the land plot. There is still a big gap between technology and architecture, architects can harness the power of technology and create design, which were never even thought it could be done before. This project is an attempt to bridge people activity within workspace with AI, which is a field used in almost all other industries but not in architecture. If such connection is reached, it will open up various possibilities that its influence will not be limited to architecture but also in how we use products. The connectivity that is now available in the contemporary world, can be used to merge different disciplines together that might make people radically change their way of living.
Suppositions: The main focus of this thesis will be about the applications of AI in the architecture realm. There will be experiments done using the AI algorithms, but the main research will be on the effect of such system on architecture, and will not be focusing on the algorithm used in much detail. Also the different aspects of AI, such as supervised learning, unsupervised learning, black box optimization, will not be researched on how to code it, but how they can be used in the project. Moreover, the thesis will not be into researching a suitable hardware for the AI system, because this needs to be scientifically researched, by engineers, and not architects, on the other hand all what comes to building materials, and how the building materials can interact with such system will be the focus of the thesis. The focus will be more on exploring different possibilities of such technology, to examine how it can play a major role in affecting the users, and the spatial formations of the building, and how it can influence the surrounding community, that can lead into a radical different way of thinking into how we live in the contemporary world.
4
Introduction
05 NEURAL NETWORKS
Theory about neural networks, underlying the most commonly used strategies, and a detailed overview about supervised vs unsupervised learning.
Introduction: Biological Neural Networks: The human brain is considered a biological neural network. It consists of interconnected neurons transmitting all time numerous patterns of electrical signals. These signal are driven by axons and received by dendrites. The decision of firing a signal or not is a decision the neuron take. Combining 100 billion neurons in one brain, each making a small decision, but over all they make a global intelligent behavior.
Artificial Neural Networks:
Human Neuron
Computer scientists have long been inspired by the human brain. In 1943, Warren S. McCulloch, a neuroscientist, and Walter Pitts, a logician, developed the first conceptual model of an artificial neural network. In their paper, “A logical calculus of the ideas imminent in nervous activity,� they describe the concept of a neuron, a single cell living in a network of cells that receives inputs, processes those inputs, and generates an output. Their work, and the work of many scientists and researchers that followed, was not meant to accurately describe how the biological brain works. Rather, an artificial neural network was designed as a computational model based on the brain to solve certain kinds of problems.There are tasks that are very sim5
ple for human and very hard computer and vise versa. A square root of 964,324, for example. A quick line of code produces the value 982, a number Processing computed in less than a millisecond. There are, on the other hand, problems that are incredibly simple for human to solve, but not so easy for a computer. Show any toddler a picture of a kitten or puppy and they’ll be able to tell you very quickly which one is which. Say hello and shake hand with someone one morning and you should be able to pick him out of a crowd of people the next day. The most common application of neural networks in computing today is to perform one of these “easy-for-a-human, difficult-for-a-machine” tasks, often referred to as pattern recognition. Applications range from optical character recognition (turning printed or handwritten scans into digital text) to facial recognition. A neural network is a “connectionist” computational system. The computational system are procedural; a program starts at the first line of code, executes it, and goes on to the next, following instructions in a linear fashion. A true neural network does not follow a linear path. Rather, information is processed collectively, in parallel throughout a network of nodes (the nodes, in this case, being neurons). The individual elements of the network, the neurons, are simple. They read an input, process it, and generate an output. A network of many neurons, however, can exhibit incredibly rich and intelligent behaviors. One of the key elements of a neural network is its ability to learn. A neural network is not just a complex system, but a complex adaptive system, meaning it can change its internal structure based on the information flowing through it. Typically, this is achieved through the adjusting of weights. In the diagram below, each line represents a connection between two neurons and indicates the pathway for the flow of information. Each connection has a weight, a number that controls the signal between the two neurons. If the network generates a “good” output, there is no need to adjust the weights. However, if the network generates a “poor” output—an error, so to speak—then the system adapts, altering the weights in order to improve subsequent results.
Neural Network 6
Neural Networks
There are numerous of strategies for making algorithm of learning, but the most common ones are :
Supervised Learning The supervised learning technique, have to involve a teacher in the process. The teacher is already smarter than the system, he gives correct answers to certain situations that the system will face. The system will make the guess, then check it with the answers given by the teacher, make a comparison between the guessed and the correct one. The system the make adjustments in its algorithm based on the errors it gets , the adjustments are continuous until the error between the guessed and the correct is minimum.
Character Recognition
Unsupervised Learning Required when there isn’t an example data set with known answers. Imagine searching for a hidden pattern in a data set. An application of this is clustering, i.e. dividing a set of elements into groups according to some unknown pattern.
Reinforcement Learning
Clustering
A strategy built on observation. Like a little mouse running through a maze. If it turns left, it gets a piece of cheese; if it turns right, it receives a little shock. Presumably, the mouse will learn over time to turn left. Its neural network makes a decision with an outcome (turn left or right) and observes its environment. If the observation is negative, the network can adjust its weights in order to make a different decision the next time. Reinforcement learning is common in robotics. At time t, the robot performs a task and observes the results. Did it crash into a wall or fall off a table? Or is it unharmed? Reinforced Learning
Neural Networks
7
Most Common Applications Pattern Recognition Examples are facial recognition, optical character recognition, etc.
Time Series Prediction: Neural networks can be used to make predictions. Will the stock rise or fall tomorrow? Will it rain or be sunny?
Signal Processing Cochlear implants and hearing aids need to filter out unnecessary noise and amplify the important sounds. Neural networks can be trained to process an audio signal and filter it appropriately. Control: You may have read about recent research advances in self-driving cars. Neural networks are often used to manage steering decisions of physical vehicles or simulated ones.
Soft Sensors A soft sensor refers to the process of analyzing a collection of many measurements. A thermometer can tell you the temperature of the air, but what if you also knew the humidity, barometric pressure, dew point, air quality, air density, etc.? Neural networks can be employed to process the input data from many individual sensors and evaluate them as a whole.
Anomaly Detection Because neural networks are so good at recognizing patterns, they can also be trained to generate an output when something occurs that doesn’t fit the pattern. Think of a neural network monitoring someone’s daily routine over a long period of time. After learning the patterns of the behavior, it could alert you when something is amiss.
8
Neural Networks
Perceptron Logic The Perceptron Invented in 1957 by Frank Rosenblatt at the Cornell Aeronautical Laboratory, a perceptron is the simplest neural network possible: a computational model of a single neuron. A perceptron consists of one or more inputs, a processor, and a single output. A perceptron follows the “feed-forward� model, meaning inputs are sent into the neuron, are processed, and result in an output. In the diagram below, this means the network (one neuron) reads from left to right: inputs come in, output goes out.
Input 0 Processor
Output
Input 1 A Single Perceptron
Step 1 Receive Inputs Lets assume there are two inputs X0, and X1. X0 = 12 X1 = 4
Neural Networks
9
Step 2 Weight Inputs Each input that is sent into the neuron must first be weighted, i.e. multiplied by some value , often a number between -1 and 1. When creating a perceptron, let’s give the inputs the following random weights: W1 = 0.5 W2 = -1 Each input is taken and multiplied it by its weight. Input 0 * Weight 0 = 12 * 0.5 = 6 Input 1 * Weight 1 = 4 * -1 = -4
Step 3 Sum Inputs The weighted inputs are then summed. Sum = 6 + -4 = 2
Step 4 Generate Output The output of a perceptron is generated by passing that sum through an activation function. In the case of a simple binary output, the activation function is what tells the perceptron whether to “fire” or not. Activation function can get really complex, leading to a strong knowledge of calculus. Therefore for the simplicity the activation function the sign of the sum. In other words, if the sum is a positive number, the output is 1; if it is negative, the output is -1. Output = sign(sum) = sign(2) = +1
10
Neural Networks
Pattern Recognition Example Task The task of this example is how to think about the logic of making a simple learning task, which is making a program to learn figuring out the equation of a straight line. A perceptron has 2 inputs (the x- and y-coordinates of a point). Using a sign activation function, the output will either be -1 or 1. The input data is classified according to the sign of the output. In the above diagram, we can see how each point is either below the line (-1) or above (+1). To avoid the confusion of point (0,0) a bias will be also added to the perceptron. The following are the main steps of its logic. • Provide the perceptron with inputs for which there is a known answer. • Ask the perceptron to guess an answer. • Compute the error. (Did it get the answer right or wrong?) • Adjust all the weights according to the error. • Return to Step 1 and repeat!
Input x Weight
Output
(X,Y)
Guess ??
Input x Weight
Bias = 1 A Single Perceptron with bias
Neural Networks
11
How to define the perceptron’s error? And how should the weights adjust according to this error? The perceptron’s error can be defined as the difference between the desired answer and its guess. • ERROR = DESIRED OUTPUT - GUESS OUTPUT In the case of the perceptron, the output has only two possible values: +1 or -1. This means there are only three possible errors. Desired
Guess
Error
-1 -1 +1
-1 +1 -1
0 -2 +2
+1
+1
0
Table Showing all ranges of Error
If the perceptron guesses the correct answer, then the guess equals the desired output and the error is 0. If the correct answer is -1 and it guessed +1, then the error is -2. If the correct answer is +1 and it guessed -1, then the error is +2. The error is the determining factor in how the perceptron’s weights should be adjusted. For any given weight, what we are looking to calculate is the change in weight, often called Δweight. • NEW WEIGHT = WEIGHT + ΔWEIGHT • Δweight is calculated as the error multiplied by the input. • ΔWEIGHT = ERROR * INPUT • NEW WEIGHT = WEIGHT + ERROR * INPUT The neural network will employ also a learning constant, that determines the rate of learning. • NEW WEIGHT = WEIGHT + ERROR * INPUT * LEARNING CONSTANT Notice that a high learning constant means the weight will change more drastically. This may help us arrive at a solution more quickly, but with such large changes in weight it’s possible we will overshoot the optimal weights. With a small learning constant, the weights will be adjusted slowly, requiring more training time but allowing the network to make very small adjustments that could improve the network’s overall accuracy. 12
Neural Networks
Equation of Straight line Simulation The following diagrams was part of a simulation done on processing software , which runs based on java language. The equation of the line is y=0.4x+1 represented in the gray line , using the above described technique the software tries to identify this equation and represented in the red line.
Equation of Straight Line Simulation
Neural Networks 13
Network Logic The Perceptron The power of neural networks comes in the networking itself, by joining many perceptrons together. Perceptrons are, incredibly limited in their abilities. A perceptron can only solve linearly separable problems. The figure Below shows a classic linearly separable data. Graph all of the possibilities; if it is possible classify the data with a straight line, then it is linearly separable. On the right, however, is non-linearly separable data. You can’t draw a straight line to separate the black dots from the gray ones.
Linear vs Non Linear Separable Problems
14
Neural Networks
Nearest Neighbors
Linear SVM
RBF SVM
Different Techniques for non linear problems (achieved by Scikit library on Python)
Neural Networks 15
One of the simplest examples of a non-linearly separable problem is XOR, or “exclusive or.” The most known separable models are the AND and the OR models. For A AND B to be true, both A and B must be true. With OR, either A or B can be true for A OR B to evaluate as true. These are both linearly separable problems. XOR is the equivalent of OR and NOT AND. In other words, A XOR B only evaluates to true if one of them is true. If both are false or both are true, then we get false. Take a look at the following truth table. It is not possible to draw a line to separate the true outputs from the false ones? Therefore perceptrons can’t even solve something as simple as XOR. But what if we made a network out of two perceptrons? If one perceptron can solve OR and one perceptron can solve NOT AND, then two perceptrons combined can solve XOR.
“AND”, ”OR” , “XOR” Logic 16 Neural Networks
The diagram below is known as a multi-layered perceptron, a network of many neurons. Some are input neurons and receive the inputs, some are part of what’s called a “hidden� layer (as they are connected to neither the inputs nor the outputs of the network directly), and then there are the output neurons, from which we read the results. Training these networks is much more complicated. With the simple perceptron, we could easily evaluate how to change the weights according to the error. But here there are so many different connections, each in a different layer of the network. How does one know how much each neuron or connection contributed to the overall error of the network? The solution to optimizing weights of a multi-layered network is known as back-propagation. The output of the network is generated in the same manner as a perceptron. The inputs multiplied by the weights are summed and fed forward through the network. The difference here is that they pass through additional layers of neurons before reaching the output. Training the network meaning adjusting the weights also involves taking the error (desired result - guess). The error, however, must be fed backwards through the network. The final error ultimately adjusts the weights of all the connections. Back-propagation need an activation function called Sigmoid function, but its details will not be discussed in this thesis, since it requires a considerable amount of calculus knowledge to understand it.
Multi Layer Neural Network
Neural Networks 17
Mesh Fitter Example The following diagram was part of a simulation done on crow plug-in on Grasshopper. It represents a self organizing nonlinear behavior known as 2D Kohonen networks. The simulation runs through a series of iterations until achieved the most accurate results.
Mesh Fitting Simulation
18
Neural Networks
19
Adaptive Environment References This chapter, has an overview of selected projects done in the scheme of adaptable environment
Lyrical Theatre in Cagliari One of the early avant-garde adaptable projects was done by the architect Maurizio Sacripanti, won the second prize in designing the new lyrical theatre in Cagliari in 1965. The theatre composed of movable plynths elements, allowing for almost infinite possibilities of stage configurations, seating layout and acoustic performance of the scenic space.
Section through the space 19
An inside view of the theater 20 Adaptive Environment References
Shape-shifting sofa Design Concept
Designed by architect Italian Carlo Ratti. It consist of a modular seats that can adjusted by hand gestures, via an app creating multiple seating configurations. It has been claimed to be the world’s first Internet-of-Things sofa. The Internet of Things is made up of smart products that communicate with each other over Wi-Fi, and can be controlled and monitored by digital devices such as smart phones and tablets.
Module Module is made up of upholstered hexagonal seats that can be used alone as stools or tessellated together into larger sofas. When the user hovers their hand over the seat, they can raise or lower it with the same movement. Placed in different configurations, the stools can be adjusted to form chairs, chaise longues and sofas. The system also gets “bored�, and begins shape-shifting on its own to engage user.
Adaptive Technique Each stool includes a mechanism that lifts the seat up and down its central stand. This motion can be controlled either with a smart-phone app, or by activating embedded sensors that recognize gestures
Adaptive Environment References 21
Above: Different Seating Configurations 22 Adaptive Environment References
Breathing Wall Design Concept
This installation is an attempt to address these question through the design of an interactive kinetic wall. One of the main contributions of this work is to explore how a physical environment can change its shape in response to hand gestures movement.
Control it explores the potential for a gesture-based interaction with our dynamic architectural space through the use of a Leap Motion device. Techniques as swiping, clicking, and dragging, as a natural, intuitive mechanism of control.
Adaptive Technique The installation consists of wood, stretchable fabric and PVC pipes controlled with Arduino micro-controller connected to a Leap Motion. The Leap Motion recognizes specific gestures, which will control several DC motors to operate several types of movement into the surface.
Adaptive Environment References 23
Above: Different Wall Configurations 24 Adaptive Environment References
Auora
Design Concept Aurora is an interactive kinetic ceiling that responds to corporeal movement below it. It examines how could a building be dynamic and develop an understanding of its users through their movements and respond accordingly, and how can we use human bodily movement as a means of interacting with a man-made artificial environment
Module This project includes five floating motion disks. The skin is made of industrial felt cut from a single 2D sheet in order to provide enough flexibility for transformations. The ‘cut’ of the industrial felt combined with the ‘drape’ of the material therefore offer the material different qualities depending on direction of the movement.
Adaptive Technique The disks can move along the z-axis but also have the ability to rotate in various direction in a 3D vector space.
Adaptive Environment References 25
Above: User Movement Detection 26 Adaptive Environment References
Bloom
Design Concept The design of the project, based on research by Sung and Wahlroos-Ritter, explores the possibilities of a thermally responsive metal surface which reacts to both the change in temperature and direct solar radiation. The team investigate the possible forms that the materials would allow whilst maintaining their conversation with broader issues of light, heat and urban space.
Module The system composed of hyper panels that sits on a foundation of 1/2� thick slumped glass panels that provide the necessary structural stability.
Adaptive Technique When the temperature of the metal is cool, the surface will appear as a solid object, once the afternoon heat penetrates the metal, the panels of custom woven bi-metal will adjust and fan out to allow air flow and increase shade potential.
Adaptive Environment References 27
Above: PAvilion at full scale 28 Adaptive Environment References
29 Thesis Statement
A detailed explanation of the system to be designed that will incubates adaptation and self learning/
Proposal The design proposal will be an interactive interior skin, placed inside a workspace, in this thesis it will examine an educational workspace. The skin will embed a self learning system that will be able to identify user behavior in terms of social behavior and in terms of physical change happening inside this working space. The system will cluster such behavior into different activities and store them into it’ knowledge base. Then the system will use such knowledge learned to respond instantaneously to any similar social behavior with a dynamic interior change that will match such behavior according to its knowledge base. The system consists of a learning phase and a self autonomous phase, but it will continuously learns from its users. As for the learning phase it is mainly about the input from user in terms of social and physical input. Regarding the social input the space will respond dynamically with users, it is as if the users are sculpting the space. This will be achieved through actuators placed in the space to give it the dynamic movement needed, and with motion tracking hand gestures that can respond to users. Allowing the space to very customized an sculpted according to the users. The system will store such input, and using unsupervised learning, and specifically using clustering technique (explained in details in Neural Networks chapter) to be able to identify activities occurring inside the space like reading, working on PC etc. Regarding the self autonomous phase, it is when the system has already been taught a number of activities happing inside the workspace. The system can then respond to a similar social behavior happened before using supervised learning and specifically pattern recognition strategy (explained in details in Neural Networks chapter), with a corresponding physical change that matches such behavior. The aim of such system is not towards optimization but to make a self reflecting, intimate and playful space. The diagram on the left elaborates all the system phases, with all the relations happening to it.
29
Diagram of main Design Concept 30 Design Proposal
31
DESIGN PROPOSAL This chapter, includes the design proposal that can combines adaptable space with self learning behavior
Case Study The case study that will be examined is an architectural studio class room in Dessau Institute of Architecture. This studio room contains approximately 15 new international student all with different needs and desires and a different point of view of how an working classroom should be like. This is why I think this richness of diversity of the school is a suitable example to start implementing the design on it and to be as my case study. The room size is 13.5 x 7.5 m and hight of 3.3 m, and it is located on the first floor of the university building.
Above Interior Pictures of the Studio 31
Plan Showing current seating arrangement and capacity 32 Design Proposal
Section Through the room
Design Proposal 33
Circulation Analysis The space has only one door 1 m in width. It contains as well numerous storage spaces placed in the middle of the room,which results for less area for circulation, and this concludes to a feeling of a denser and not comfortable space with very inefficient circulation.
Circulation inside the room 34 Design Proposal
Space Dynamic Elements This analysis shows all the dynamic components inside the room. There flexibility of components varies, some are very flexible like tables and chairs, and some are much flexible like storage area. These components are questioned to be replaced by an interior skin that can perform all their function, but in a more adaptable way but more important in a more understanding way of the user needs.
Dynamic Components
Design Proposal 35
Not all the dynamic components will be replaced, only those which are not in the reach of users easily and those which are not flexible enough to change. This conclude that seating and furniture will not be replaced by the skin. Since it is something the users can change easily and it is totally in their reach. On the other hand, other components that controls illumination, privacy, noise, and area, which is not always in the reach of the users, these components will challenged to be replaced by the skin.
Flexible
Semi Flexible
Flexibility Analysis
36 Design Proposal
Static
Materials Proposed Soft and Strong When searching of a skin that can be adaptable, it should be also considered that it is should be strong as well. A similar example can be found the medieval chain mail armor, where they were a kind of mesh that is smooth and fits body perfectly but at the same time very strong and durable. From such flexible and strong system, it was the inspiration to imitate this behavior in the proposed skin. It will be mainly consisting of a metallic mesh from the outside to give the flexibility and strength, and from inside will be a smooth felt fabric, that gives the coziness and warmth of the space.
Metallic Flexible Armour
Design Proposal 37
Inflatable Structure Inflatable structures has many advantages that make it one of the suitable to use it in an adaptable environment. Theses advantages can be briefed in: • Light Weight • Cheap • The flexibility of changing shape
Inflatable Structures made by AIRFORM 01 to fill interior
38 Design Proposal
Air Forest - Mass Studies (Japan)
Learning Phase The first part of the scenario is the learning phase, the system start blind not knowing anything happening inside the space. It starts to learn from its users by their input by social behavior and by their physical change they perform inside the space. The system starts recording these inputs and using unsupervised learning clustering algorithm it can start identifying repeating activities. Activities that occurred in the case study, and which be later put into the design are : • Private Seating • Reading • B uilding a Physical Prototype • Group Presentation The figure below shows social characteristics of such activities in which the system can recognize the activities.
Presentation Configuration Presentation Presentation Configuration Configuration
Private Seating Configuration Private Seating Private Seating Configuration Configuration Electronic Interaction
ElectronicUser-User InteractionInteraction Proximity User-User Interaction Electronic Interaction Proximity User-User Interaction Users
Proximity Users Users
Electronics
Electronics
Reading ReadingConfiguration Configuration Reading Configuration
Group Group WorkWork Configuration Configuration Group Work Configuration
Electronics
Design Proposal 39
Using Hand gestures sensors technology , combined with actuators in the skin, user can start literally sculpting the space and crease the spaces that are suited to their activities. Such physical input as well with social input corresponding is being stored into the system. Hand Gestures Sensor: Leap Motion
Active Sensor Inactive Sensor
User Sculpting Space Using Hand Gesture Component Leap Motion 40 Design Proposal
Leap Motion Technology Leap Motion is an already existing hand gesture technology. The concept behind leap motion is very simple, a small gadget about the size of a smart phone is connected to a pc and placed in front of it, within an eight-cubic-foot cone of space above it, the controller can track motions as small as .01 millimeters, capturing the gestures and translating them into the computer according to the purpose for which it is used.
Leap Motion Technology
Design Proposal
41
Leap Motion Integration with Skin The following simulation was done by connecting leap motion with grasshopper, as a test to make an interactive interface that can be done in reality if was connected by physical actuators that can control the skin. There are two main functions done in the simulation , the leap motion sensor detects the hand height and adjust accordingly the height of the skin. It can also detects if hand is closed or open and simulate a similar effect to openings. There can be also, more hand gestures to be embedded since the leap motion can detect very accurately hand gestures.
Detecting Hand Height
Detecting Hand Opening
42 Design Proposal
Emerged Patterns Below are some physical elements in space that usually change when such activities occur, the figure below shows how a system can recognizes the-different activities in terms of social and physical patterns.
Physical Requirements corresponding to activities
Social Behavior corresponding to activities
Design Proposal 43
Self Autonomous Phase The Second part is the self autonomous phase, it is where the system is mature and intelligent enough to start responding to the social behavior happening inside the space, with a physical change, using supervised learning pattern recognition technique. The diagram below represent the variation of characteristics of both social or physical in terms of radius of sphere, each when put under its own activity can result in a formation of a pattern that the system can then use to respond correspondingly.
The Formation of Patterns
44 Design Proposal
Pattern Recognition Tool To run the simulation of the pattern recognition supervised learning in the self autonomous phase, a plug-in on grasshopper was used called Crow. It can implement an existing engine for artificial neural networks, called NeuronDotNet3.0 developed by Vijeth Dinesha in the Grasshopper environment. The plug-in performs the reinterpretation of input vectors into output vectors according to previously defined training samples. It uses a classic back-propagation network. The network structure can be defined by providing inputs for number and type of network layers, the number of neurons in each layer, a learning rate and the number of training cycles to be run previously to the actual input vector translation. The steps for running the definition runs as follow: • Defining Number of training cycles. • Specify network layer type ( Sigmoid, linear, logarithmic etc. As mentioned in the supposition , there will not be a detailed explanation in coding these functions. • Identify number of neurons per layer. • Create input pattern in the form of a matrix, for example in the example studies : 3 activities X 4 social behavior characteristics. Therefor, it is needed a 4x3 matrix for the input phase. • Create output pattern in the form of matrix, in the example studied. • Each new Input is then compared with with predefined ones • According to similarity of the pattern the code can classify, in which category the pattern falls.
Design Proposal 45
Pattern Recognition Simulation The below diagrams are part of pattern recognition simulation done using crow plug-in as mentioned in the previous pages, using pattern recognition technique to identify which pattern falls in what category.
Identifying Pattern 1
46 Design Proposal
Identifying Pattern 2
Using same technique to convert the output instead of just circles, but can be an emerged physical configuration, in the test dune it was to simulate : • Private Configuration • Pair Configuration • Group Configuration
Private Configuration
Group Configuration
Pair Configuration
Design Proposal 47
Modules Used Their are mainly two modules controlling the interior space. The first one is controls the sizes of the zones in terms of height and area. Using Metallic telescopic pistons fixed to actuators nodes that are distributed among the allowed imaginary box frame of the class room. The second module is a sub module from the first one, which acts a control for daylight. Both modules can be totally controlled by the users of the space. The only limitation that the users have that the module controlling the volume should not exceed the volume of the box frame, which was originally the volume of the room.
Controlling Volumes using air pumped Pistons
48 Design Proposal
Controlling Daylight Units
Design Proposal 49
Inflated Partition For Controlling Privacy 50 Design Proposal
Inflatable Part
Emerged Configurations The following drawings shows a proposal of an emerged design after the system being self autonomous and responding to the different functions activities as shown in the flowing diagram.
Activities Configuration
Design Proposal 51
Emerged Design 52 Design Proposal
Plan View 2.2m Level
Design Proposal 53
Section View 1
54 Design Proposal
Section View 2
Design Proposal 55
Elevation 1
Elevation 2 56 Design Proposal
Elevation 3
Elevation 4
Design Proposal 57
Configuration 1
Configuration 2
Configuration 3
58 Design Proposal
Configuration 1
Configuration 2
Configuration 3
Design Proposal 59
Exterior Shot
Interior Shot 2
64 Conclusion
A self critique on the proposed design, how can system be further implemented, and how it can affect people life style
Proposed Structure One Module cannot fit everything It can be concluded that one module cannot perform all functions inside the space. One technique like origami can be very useful when dealing with controlling daylight, but not so useful when dealing with furniture design. Pistons, on the other hand can very practical in making refined extrusions inside the space, but is not practical to make any openings. Thus, it is more efficient to create a module that is optimized for it adaptable function. The challenge if the are multiple modules in one space, how they can still look coherent and similar, yet performing different functions.
No need to make everything adaptable When thinking of making an adaptable space, it should be also thought of what is really to be adaptable and what is not. If all the component inside the space, were just changed to be adaptable just for the sake of it, then it might end up being over expensive and very complex that reach to a point where it is inefficient, since very complex systems, needs maintenance, and energy. There fore it is best to consider to choose what to make adaptable and what not to make. In the case study examined, it was not about making the already very flexible furniture adaptable, but making things that are not from the users reach adaptable. These elements are, privacy, illumination, height, and area. Such elements are not within the the control of the user easily and therefore needed to be adaptable.
Further Implementation From Interior to architectural to urban level This thesis just examined how an adaptable workspace can be like jut from the interior pint of view, not regarding the surrounding workspaces on the architectural level. A further implementation of this thesis is start thinking when there are multiple workspaces next to each other, how will they will co-
64
herently adapt together, and will there be a kind of negotiation of spaces on an architectural level. Can one workspace take some volume of another workspaces, and can be there such system intelligent enough that can learn from the user needs but still keep balance of the whole society, who will control or have the full access to such system, such question can be challenging to dig to which will open up many potentials and concerns at same time.
Effect on Life style Space Currency Such model will open up also of redefining of what is called your own private space. In the contemporary world space is defined by physical border. This border is usually a wall that separates spaces and define public and private space. But on a fully adaptable environment where space can change all the time, it becomes different to determine what are the users border, and how to measure it, can it be by that each user has a certain amount of volume that can change its shape but maintain total volume, but then as well there different activities which requires different volumes, which might lead also to an inefficient design. Unless there is a system that can clearly defines borders and who can control it, users might also even sacrifice the luxury of having an adaptable space in the sake of having clear borders that can define their own space.
Laziness A fully adaptable system, where everything is perfectly fitted ergonomically for the users comfort, can lead also to laziness. People do not need to walk to other places that have different function, but such functions can come to them all in room. If giving such freedom of changing the space, this could eventually lead to an unhealthy lifestyle if was to misused.
Death of History History is manly defined by its architecture, last for centuries can make people have look of what the world looked like back then, and it give an identity to the space. On the other hand, even if structures were huge in scale or very complex, they will always be temporary and can leave no trace of what looked like. Can architecture history be affected with such system. Is it possible shape people identity in a fully adaptive environment.
Conclusion 65
Artificial Intelligence Influence Offloading Cognition Is it possible that artificial intelligence, algorithms can make people offload all the tasks they do not waste time in designing it to machines. This will might open up that people can be free to live in spaces and places of imagination.
“The future of humanity is us moving into the imagination, to fully turn our minds inside out, and inhabit spaces that are linguistic constructs, spaces built and shaped by mind and thought and intentionality. � Terence McKenna
66 Conclusion
67 References • Scikit-learn. (n.d.). Retrieved June 14, 2016, from http://scikit-learn.org/stable/ • Behnaz Farahi. (n.d.). Retrieved June 14, 2016, from http://behnazfarahi.com/ • Crow | felbrich.com. (n.d.). Retrieved June 14, 2016, from http://www.felbrich.com/projects/Crow/Crow.html • Douglis, E. (2009). Autogenic structures. New York: Taylor & Francis. • Lima, M. (2011). Visual complexity: Mapping patterns of information. New York: Princeton Architectural Press. • Maas, W., Hackauf, U., Ravon, A., & Healy, P. (n.d.). Barba: Life in the fully adaptable environment. • A Step by Step Backpropagation Example. (2015). Retrieved June 14, 2016, from https://mattmazur.com/2015/03/17/ a-step-by-step-backpropagation-example/ • Parametricism.co.uk. (n.d.). Retrieved June 14, 2016, from http://www.parametricism.co.uk/blog/portfolio/dodo-hasreborn-a-new-plugin-for-grasshopper/ • Pasquinelli, M. (2015). Alleys of your mind: Augmented intelligence and its traumas. Lüneburg: Meson Press, Hybrid Publishing Lab, Centre for Digital Cultures, Leuphana Univeristy of Lüneburg. • Welcome to PyBrain. (n.d.). Retrieved June 14, 2016, from http://pybrain.org/ • Richert, W., & Coelho, L. P. (n.d.). Building machine learning systems with Python: Master the art of machine learning with Python and build effective machine learning sytems with this intensive hands-on guide. • Russell, S. J., & Norvig, P. (1995). Artificial intelligence: A modern approach. Englewood Cliffs, NJ: Prentice Hall. • Lyrical Theatre in Cagliari, Maurizio Sacripanti, 1965 – – SOCKS. (2012). Retrieved June 14, 2016, from http:// socks-studio.com/2012/06/10/lyrical-theatre-in-cagliari-maurizio-sacripanti-1965/ • New shape-shifting sofa will let you sit down anywhere. (2016). Retrieved June 14, 2016, from http://www.sciencedump.com/content/new-shape-shifting-sofa-will-let-you-sit-down-anywhere • Shiffman, D., Fry, S., & Marsh, Z. (n.d.). The nature of code. • Spyropoulos, T., Steele, B. D., Holland, J. H., Dillon, R., Claypool, M., Frazer, J., . . . Burry, M. (2013). Adaptive ecologies: Correlated systems of living. London: Architectural Association. • Tedeschi, A., Wirz, F., & Andreani, S. (n.d.). AAD, Algorithms-aided design: Parametric strategies using Grasshopper.
67