LEE- Mimari Tasarımda Bilişim-Yüksek Lisans
Bu koleksiyon için kalıcı URI
Gözat
Konu "yapay zeka" ile LEE- Mimari Tasarımda Bilişim-Yüksek Lisans'a göz atma
Sayfa başına sonuç
Sıralama Seçenekleri
-
ÖgeRizobot: Collective form finding through swarm robotics(Graduate School, 2022-12-16) Balcı, Ozan ; Alaçam, Sema ; 523201010 ; Architectural Design ComputingThis study proposes a framework for performing form-finding studies using a swarm of mobile robots. During the development process of the proposed framework, 3 different case studies consisting of 5 experiments were conducted in order to observe the effect of different features on the capability of the swarm system. The term RIZOBot is proposed by the author. Two types of robots named RIZOBot-Mini and RIZOBot were used as agents in the case studies. Both of these robots are low-fidelity differential-drive mobile robots developed as part of this research. RIZOBot-Mini is a wheeled robot which is smaller and faster compared to RIZOBot, while RIZOBot is a tracked robot, consisting of a larger body and more powerful motors. The robots in question have a hardware and software infrastructure that can be equipped and operated with different actuators to conduct various research on swarm systems. In the first case study of the research, the potential of swarm robotics in experimental artwork was examined. In this context, 4 RIZOBot-Minis equipped with different coloured inks to perform drip painting were used as agents. Operating with a semi-central system, the robots took a predefined trigonometric equation as a defined route and manipulated that route by interacting with each other. Robots moving on a bordered white canvas, left their traces on the canvas by drip painting during the experiment. At the end of the experiment, the effect of the robot-robot interaction feature on the swarm was observed through these traces. In the second case study, form studies were carried out by drawing action on a swarm of zone-sensitive mobile robots. In the proposed system, 4 RIZOBot-Minis equipped with different coloured markers were used as agents. The robots sensed the light intensity at their location with the light sensor they have and exhibited different movement behaviors depending on the measurement results. In the study consisting of two experiments, the first experiment focused on the robot-environment feature, while the second experiment proceeded through robot-robot communication. In the first experiment, robots moving in an area with light and dark zones demonstrated two different movement patterns according to the area they were in. In the second experiment, the swarm aims to find the brightest spot in the area by communicating with each other. Robots constantly perform light sensing and compare their measurements with each other. The robot with the highest reading oscillates around itself in the same location, while the rest of the swarm continues to search for a brighter spot. Thus, agents which do not have any localization feature, collectively find the brightest spot in the given area. In both experiments, the robots drew their traces on a white canvas with markers. After the experiments, these traces were examined and the behavior of the swarm of RIZOBot-Minis was observed and analyzed. In the last case study of the research, form-finding studies were carried out in an outdoor environment using a swarm of 4 robots which perform adding/pouring action. In the study, 4 RIZOBots, each of them having a tank full of granulated sugar and anozzle that can be opened and closed, were used as robots. The study consists of two experiments. In the first experiment, the robots aim to find one light source placed on the sand floor in the given area and gather around it. The first RIZOBot that finds the light source terminates its movement, broadcasts infrared signals from its transmitter unit and calls the rest of the swarm. Learning that the light source is found by another robot, the rest of the swarm follow the transmitted signals with their receiver unit by using the localization feature and aim to reach the light source. Meanwhile, the RIZOBots following the signal open their nozzles and pour sugar on the ground during their movements. Each RIZOBot that reaches the light source imitates the first robot that finds the light and amplifies the emitted signal. In the experiment focusing on the robot localization feature, the experiment ends when the entire swarm reaches the light source. In the second study, two light sources are placed on the sand floor and RIZOBots seek these light sources. After the 2 robots from the swarm find these sources, they emit infrared signals as in the first experiment. The rest of the swarm randomly selects one of the two robots that find the light and follows its signals to reach it. RIZOBots, reaching one of the robots that find the light, takes the other robot as a new target, and moves towards it by pouring sugar on the sand. Changing their destination with the other robot as they arrive at each one, RIZOBots constantly move between the two light sources and leave their traces on the ground. The experiment ends when the sugar in the tank of the robots runs out. Robots, which distinguish two different signals with the robot-robot recognition feature, create a 2.5-dimensional form on the sandy ground with the movement traces they leave between two light sources. At the end of the two experiments, the forms created on the sandy ground were documented and analyzed. In the proof-of-concept study, which consists of the aforementioned 3 case studies, the form-finding potential of a swarm of mobile robots is examined through certain features with both swarm and hybrid control architectures, and a framework is proposed in line with these examinations. Preliminary results show that the proposed framework enables indirect user-swarm interaction and has the potential to act as a co-designer rather than just a tool in the early phases of architectural design.
-
ÖgeYeni fenomen algoritmalar: çekişmeli üretken ağların mimarlıktaki potansiyelleri üzerine bir araştırma(Lisansüstü Eğitim Enstitüsü, 2021-07-13) Eroğlu, Ruşen ; Gül, Leman Figen ; 523171010 ; Mimari Tasarımda Bilişim ; Informatics in Architectural DesignGün geçtikçe gelişen yapay zeka, hızla geniş bir araştırma alanına dönüşmektedir. Yapılan tez çalışması, işte tam bu noktada; son yıllarda oldukça gündemde olan yapay zeka yöntemlerini kullanarak mimari görsel üretim özelinde potansiyelleri araştıracaktır. Bununla birlikte, çalışmanın bir diğer vurgusu üretken modelleri eğitme sürecinde bilgisayarın görüntüleri nasıl anladığı ve oluşturduğunu anlamamızı sağlayacağıdır. Bilgisayarın nasıl gördüğünü anlamak, bunun potansiyellerini keşfetmek; bizi imaj üretiminin bir adım ötesine taşıyacaktır. Bu şekilde bu sistemlerin gelişmesinde rol alma olanağı verecektir. Tez, sırasıyla; veri biliminin Çekişmeli Üretken Ağlara kadar olan gelişimini açıklamakla birlikte bu gelişmelerin mimarlık disiplinindeki etkilerini anlatmaktadır. Çekişmeli Üretken Ağlar ile mimarlık alanında yapılmış çalışmaların açıklandığı literatürde kullanılan modellerden çeşitli Çekişmeli Üretken Ağların deney için seçilmesine karar verilmiştir. Farklı derecede özniteliklere sahip veri setlerinin, yapısal ve üretim döngüleri ile seçilen dört Çekişmeli Üretken Ağ, tezin ana bölümünü oluşturmaktadır. Bu ağlar ile yapılacak deneyler için kullanılan araçlar; Google Colab bulut ortamı ve Anaconda uygulamasındaki Jupyter Notebook arayüzünde Python programlama dili olarak seçilmiştir. Seçimde kullanılan modellerin bu programlama diline uygunluğu göz önüne alınmıştır. Sonuç olarak; deneyler sonucu üretilen imajların görsel potansiyelleri mimari perspektifte irdelenmiş, bulunan keşiflerden potansiyeller çıkarılmıştır. Bu bakımdan tezin yürütülmesi, denetimsiz bir derin öğrenme modelini andırmaktadır. Çalışma; açık kaynak paylaşımlı olan DCGAN, Pix2Pix, CycleGAN ve StyleGAN algoritmaları ile yine açık kaynak alınmış üç veri kümesi olan Ahameniş, Bauhaus ve Paladyan tarzı veri setleri çalıştırılarak başlamıştır. Deneylerde ortaya çıkan geri dönüşler ile yeni veri setleri oluşturulmuştur. Böylece deneylerde çeşitlilik sağlanmış, kontrollü değişkenler ile toplam 9 deney organize edilmiştir. Bu dokuz deneyde veri setleri öznitelik farklılıklarına ya da modelin ihtiyaçlarına göre değiştirilmiştir. Çıkan sonuçlar sezgisel olarak dolaylı nitel yöntemle değerlendirilmiştir. Mimarlık disiplinindeki tasarım ve üretim potansiyellerini arayan çalışma, bu gözlemler sonucu mimari görsel üretim anlamında hem stil transferi hem de form üretimi konularında keşifler ve bu keşiflerden potansiyeller açığa çıkarmaktadır. Sonuç olarak; araç olarak seçilen yapay zeka algoritmaları, mimarlık için yeni ilkeler, kurallar ve yollar oluşturma fırsatı vermektedir. Yapay zeka alanında bu algoritmalar gelişirken; buna açık olmak ve kullanım alanlarını ölçmek yeni bir düşünsel bakış açısı getirebilir.