RESEARCH REPORTS Institute for Computational Design and Construction Samuel Leder Co-Design of Collective Robotic Construction Systems in Architecture 18 Samuel K Leder CO-DESIGN OF COLLECTIVE ROBOTIC CONSTRUCTION SYSTEMS IN ARCHITECTURE RESEARCH REPORTS Institute for Computational Design and Construction 18 RESEARCH REPORTS Institute for Computational Design and Construction Edited by Prof. Achim Menges, AADipl(Hons) Samuel K Leder CO-DESIGN OF COLLECTIVE ROBOTIC CONSTRUCTION SYSTEMS IN ARCHITECTURE © 2025 Institute for Computational Design and Construction University of Stuttgart Keplerstrasse 11 70174 Stuttgart Germany D 93 RESEARCH REPORTS Institute for Computational Design and Construction 18 ISBN 978-3-9824862-7-7 All rights, in particular those of translation, remain reserved. Duplication of any kind, even in extracts, is not permitted. The publisher has no responsibility for the continued existence or accuracy of URLs for external or third-party internet websites referred to in this book, and does not guarantee that any content on such websites is, or will remain, accurate or appropriate. Foreword Collective robotic construction offers a radically different approach to current construction processes, in which several distributed, mobile and agile robot units collectively construct complex architectural structures from a multitude of simple elements, which also challenges conventional approaches to architectural design. Samuel Leder’s dissertation examines a co-design approach to collective robotic construction, which is characterized by a novel system in which wooden building elements are first used as part of the robot system and then integrated into the construction system to be assembled. This approach also requires highly integrative architectural design methods, which are the focus of Samuel Leder’s scientific contribution. His work impressively demonstrates how original disciplinary insights in the field of architectural design methods can be gained by genuinely interdisciplinary research. Through a high degree of methodological rigour and competence, Samuel Leder has led a risky research project to an excellent result! Prof. Achim Menges, AADipl(Hons) CO-DESIGN OF COLLECTIVE ROBOTIC CONSTRUCTION SYSTEMS IN ARCHITECTURE A dissertation approved by the Faculty of Architecture and Urban Planning of the University of Stuttgart for the conferral of the title of Doctor of Engineering Sciences (Dr.-Ing.) Submitted by Samuel K Leder from Montclair, New Jersey, United States of America Committee Chair: Prof. Achim Menges, AADipl(Hons) Committee member: Prof. Dr.-Ing. Marc Toussaint Date of the oral examination: 29.11.2024 Institute for Computational Design and Construction of the University of Stuttgart 2025 CO-DESIGN-METHODEN FÜR KOLLEKTIVE ROBOTISCHE KONSTRUKTIONSSYSTEME IN DER ARCHITEKTUR Von der Fakultät Architektur und Stadtplanung der Universität Stuttgart zur Erlangung der Würde eines Doktor-Intenieurs (Dr.Ing.) genehmigte Abhandlung Vorgelegt von Samuel Leder aus Montclair, New Jersey, Vereinigte Staaten von Amerika Hauptberichter: Prof. Achim Menges, AADipl(Hons) Mitberichter: Prof. Dr.-Ing. Marc Toussaint Tag der mündlichen Prüfung: 29.11.2024 Institut für Computerbasiertes Entwerfen und Baufertigung der Universität Stuttgart 2025 Acknowledgements The work presented in this dissertation would not have been possible without the encouragement, guidance, and generosity of family, friends, and colleagues. At times, the lines between these roles have blurred, and I am grateful for each person who has been part of this journey in their own way. I am deeply indebted to Prof. Achim Menges for many years of mentorship. Thank you for believing in my work, taking a chance on my ideas, and fostering an environment at the Institute for Computational Design and Construction (ICD) and the Cluster of Excellence “Integrative Computational Design and Construction for Architecture” (IntCDC) that has been invaluable to my development as a researcher, academic, and individual. I would also like to express my gratitude to my second reviewer, Prof. Dr. Marc Toussaint, whose perspectives have helped shape me as an interdisciplinary researcher. Our discussions challenged me to think beyond my own boundaries and sparked new lines of inquiry. It has been a privilege to conduct this research alongside many talented members of the ICD. Thank you, Zuardin Akbar, Martín Alvarez, Dr. Felix Amtsberg, Dr. Serban Bodea, Dr. Oliver A. Bucklin, Dr. Tiffany Cheng, Niccolò Dambrosio, Anni Dai, Dr. Karola Dierichs, Diellza Elshani, Monika Göbel, Harrison Hildebrandt, Zhenxiang Huang, Fabian Kannenberg, Laura Kiesewetter, Long Nguyen, Dr. Luis Orozco, Maria Papadimitraki, Katja Rinderspacher, Gili Ron, Ekin Sila Sahin, Dr. Tobias Schwinn, Shermin Sherkat, Lasath Siriwardena, Lior Skoury, Tim Stark, Christoph Schlopschnat, David Stieler, Dr. Yasaman Tahouni, Simon Treml, Hans Jakob Wagner, Xiliu Yang, Dr. Christoph Zechmeister and Max Zorn. Special thanks to my “foreparents” in mobile robotic systems at the ICD, Dr. Maria Yablonina and Dr. Nathan Melenbrink, whose guidance laid the foundation for my work in this field. I would also thank my former students-turned-colleagues, Philipp Kragl, Nicolas Kubail Kalousdian, and Nils Opgenorth, with your collaboration, I was able to establish and expand the Distributed Robotics Research Area at the ICD. I am also especially thankful to three individuals: Hana Svatoš-Ražnjević, for being a wonderful roommate and a source of encouragement during the final writing and presentation preparation days; Rebeca Duque Estrada, for a best friendship I will always treasure; and Mathias Maierhofer, for your steadfast help and understanding in ways that only you could provide. The interdisciplinary work in this dissertation would not have been possible without the expertise, patience, and commitment of many members of the IntCDC. I am especially grateful to my direct project collaborators: Prof. Dr. Marc Toussaint, Prof. Dr. Metin Sitti, Dr. Ögüz Salih Özgür, Dr. Valentin Noah Hartmann, and Dr. Hyun Gyu Kim. I would also like to thank the additional members of the initial Early Career Board of the IntCDC, whose encouragement and advice sustained me in the early days of learning what it truly means to undertake a dissertation: Piotr Baszynski, Dr. Tiffany Cheng, Dr. Oliver Gericke, Dr. Yijie Gong, Dr. Anna Krtschil, Dr. Pascal Mindermann, Dr. Luis Orozco, Dr. Jan Petrs, Alya Rapoport, Hana Svatos-Raznjevic, Dr. Yasaman Tahouni, Dr. Janusch Töpler, and Dr. Christian Vöhringer. I am equally grateful to all the ITECH students with whom I had the privilege of working during my dissertation. Your creativity has been a constant source of inspiration. A special thank you to those who entrusted me with supervising your master’s theses: Nicolas Kubail Kalousdian, Philipp Kragl, Grzegorz Łochnicki, Matthew Johnson, Daniel Locatelli, Nils Opgenorth, Sai Praneeth Singu, Ali Shokri, Xin Sun, Michael Tucker, and Chia-Yen Wu. Above all, I am grateful to my family. Bruce, Sylvia, Jamie, and Andrew, I feel your love and encouragement across the ocean. To Ozzy and Tucker, thank you for your unwavering companionship. To my extended family on both sides and non-blood relatives [the Taub/Duckler, the Levine/Ander], I am grateful for your constant encouragement. To my friends spread across the globe, your messages, calls, and visits have been a source of joy and and a sense of grounding throughout my dissertation journey. And again to Mathias, I cannot imagine having undertaken this journey without you by my side. Samuel K Leder Contents Foreword iii Acknowledgements ix List of Abbreviations xvii List of Figures xix List of Tables xxiii Abstract xxv Zusammenfassung xxvii 1 Introduction 3 1.1 Research Aim 5 1.2 Dissertation Structure 6 2 Context and Background 9 2.1 Multiple Robotic Systems 9 2.2 Computational Design and Digital Fabrication 11 2.2.1 Co-Design in Architecture 12 3 Research Objectives and Questions 15 3.1 Co-design of CRC Systems 15 3.2 Architectural Design in CRC 16 xiii Contents 4 Current State of the Technology 19 4.1 Collective Robotic Construction 19 4.1.1 Existing Research on Architectural Design in CRC 21 4.1.1.1 ABMS 23 5 Research Structure and Methods 27 5.1 Modular CRC System for Timber Construction 28 5.1.1 Concept Overview 28 5.1.1.1 Design Space 29 5.1.1.2 Robotic System 30 5.1.1.3 Material System 32 5.1.1.4 Robotic Basic Motion Primitives & Assembly Tasks 33 5.1.1.5 Physical Experiments 35 5.1.1.6 Digital Twin/Interface 36 5.2 Agent-Based Architectural Design Methods 37 5.2.1 Justification for ABM in CRC 38 5.2.2 Conceptual Models 39 5.2.2.1 Approach: Agent represents Building Material 40 5.2.2.2 Approach: Agent represents a Mobile Robot 41 5.2.3 Software Implementation: ABxM 43 6 Publications 47 6.1 Article A: Architectural Design in Collective Robotic Construction 47 6.2 Article B: Leveraging Building Material as Part of the In-Plane Robotic Kinematic System for Collective Construction 64 6.3 Article C: Enhanced Co-design and Evaluation of a Collective Robotic Construction System for the Assembly of Large-Scale In-plane Timber Structures 82 6.4 Article D: Merging Architectural Design and Robotic Planning using Interactive Agent-Based Modelling for Collective Robotic Construction 107 6.5 Article E: Introducing Agent-Based Modeling Methods for Designing Architectural Structures with Multiple Mobile Robotic Systems 125 xiv Contents 7 Results 141 7.1 Demonstration of Assembly Tasks 141 7.1.1 General Workflow 143 7.1.2 Formal ABM for Architectural Design 144 7.2 Large-Scale In-Plane Prototype 146 7.2.1 General Workflow 148 7.2.2 Formal ABM for Architectural Design and Robotic Path Planning 149 8 Discussion 153 8.1 Co-Design in CRC 153 8.2 Architectural Design in CRC 157 8.2.1 Existing Approaches to Architectural Design in CRC 157 8.2.2 Computational Models for Architectural Design in CRC 159 8.2.3 Design Intent 160 8.2.4 CRC Workflows and Phases of Construction 160 9 Conclusion and Outlook 165 A Supporting Publications with Contribution by the Author 169 A.1 Appendix A - Publication 1 170 A.2 Appendix A - Publication 2 171 A.3 Appendix A - Publication 3 172 A.4 Appendix A - Publication 4 173 A.5 Appendix A - Publication 5 175 A.6 Appendix A - Publication 6 176 B Supporting Educational Workshops with Contribution by the Author 179 B.1 Appendix B - Workshop 1 181 B.2 Appendix B - Workshop 2 182 B.3 Appendix B - Workshop 3 184 B.4 Appendix B - Workshop 4 186 xv Contents C Supporting Thesis Projects Advised by the Author 189 C.1 Appendix C - Project 1 190 C.2 Appendix C - Project 2 191 C.3 Appendix C - Project 3 192 C.4 Appendix C - Project 4 194 C.5 Appendix C - Project 5 195 D Project Credit List 199 D.1 IntCDC RP19-1 199 D.2 IntCDC RP19-2 200 D.3 Zukunftbau 201 Bibliography 203 Image Credits 211 Curriculum Vitae 213 xvi List of Abbreviations ABM Agent-Based Model ABMS Agent-Based Modeling and Simulation ABMs Agent-Based Models AEC Architecture, Engineering and Construction AI Artificial Intelligence CAD Computer-Aided Design CAM Computer-Aided Manufacturing CRC Collective Robotic Construction DOF Degrees of Freedom IntCDC Integrative Computational Design and Construction for Architecture LGP Logic Geometric Programming MRS Multiple Robot Systems or Multi-robot Systems SDK Software Development Kit xvii List of Abbreviations TAMP Task and Motion Planning UML Unified-Modeling-Language xviii List of Figures 1.1 Collective robotic construction (CRC) system composed of timber struts and robotic actuators that are required to work together in order to assemble timber structures. . . . . . . 2 1.2 Various approaches to construction automation using different scales of robots including a large-scale building construction factory (Top Left) [68], a tower crane (Top Right) [72], an industrial robotic arm on mobile platform (Bottom Left) [12], and a team of bespoke mobile machine (Bottom Right)[57] . . . . . . . . . . . . . . . . . . . . . 5 2.1 Initial physical prototype of the robotic actuator for the CRC system discussed in this dissertation. . . . . . . . . . . . . 8 3.1 Final physical prototype of a robotic actuator with the screwdriving effector, utilized for connecting timber struts. 14 4.1 Kinematic chain composed of two robotic actuators and one timber strut. . . . . . . . . . . . . . . . . . . . . . . . . . 18 5.1 Four robotic actuators in two kinematic chains ready to start the process of assembling a large-scale in-plane assembly. . 26 5.2 Kinematic representation of a kinematic chain with two robotic actuators and one timber strut [36]. . . . . . . . . 31 xix List of Figures 5.3 Effector for assembling timber struts together (A), including diagrammatic representations of how the two major mechanisms for screwdriving and screw feeding function (B-C) [36]. . . . . . . . . . . . . . . . . . . . . . . . . . 33 6.1 Four robotic actuators in two kinematic chains in the process of assembling a large-scale in-plane assembly. . . . . . . . 46 7.1 Four robotic actuators in two kinematic chains in the process of assembling a large-scale in-plane prototype. . . . . . . . 140 8.1 Overlay of the ABM on a photograph of the physical CRC system [39]. . . . . . . . . . . . . . . . . . . . . . . . . . 152 8.2 Qualitative comparison of the proposed CRC system to existing automation in construction research. Projects in the context of CRC, as well as in timber construction, were utilized for comparison. The top row is from the proposed CRC system [36]. . . . . . . . . . . . . . . . . . . . . . 154 8.3 Quantitative comparison of the proposed CRC system to existing automation in construction research. Projects in the context of CRC, as well as in timber construction, were utilized for comparison. The top row is from the proposed CRC system [36]. . . . . . . . . . . . . . . . . . . . . . 155 8.4 Comparison of existing CRC systems based on volumetric fabrication rate, density of building element material, and the weight of a single building element. The proposed system appears twice with red dots to highlight the different fabrication rates when connection and material pick up are considered or disregarded [36]. . . . . . . . . . . . . . . 156 9.1 Further development of the discussed CRC to assembly more spatial timber assemblies. . . . . . . . . . . . . . . . . . . 164 xx List of Figures 9.2 Approaches to construction automation with robots currently explored in the IntCDC: tower cranes (Top Left), spider cranes (Top Right), industrial robotic arms (Bottom Left), and collections of custom small-scale robots (Bottom Right). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168 B.1 CRC system for the adaptation on interior spaces using fabrics whose corner points are moved around by mobile robots with magnets. . . . . . . . . . . . . . . . . . . . . 181 B.2 Visualization to inspire the conceptualization and re imagination of a CRC system composed of wheeled robots that push around digital material. . . . . . . . . . . . . . . 183 B.3 CRC system inspired by the painted desert challenge. . . . 184 B.4 Photograph of CRC system composed of wheeled mobile robots and digital material in action. . . . . . . . . . . . . 187 C.1 CRC system for bamboo construction (Łochnicki, G., and Kubail Kalousdian, N., 2020). . . . . . . . . . . . . . . . 190 C.2 Multi-scalar robotic construction system for multi-storey timber construction based on press gluing (Locatelli, D., and Opgenorth, N., 2021). . . . . . . . . . . . . . . . . . . . . 191 C.3 Differential Boundary Deposition mobile robot (Johnson, M., Kragl, P., and Tucker, M., 2022). . . . . . . . . . . . . 193 C.4 DendriBot mobile robot (Shokri, A., and Singu, S. P., 2023). 194 C.5 CRC system for the assembly and rearrangement of deployable building blocks (Sun, X., and Wu, C., 2023). . . 196 xxi List of Tables 5.1 Outline of major global design decisions for the each ABMS approach . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 7.1 Effect of the design intent parameters on the overall designed structure and assembly process in the formal ABM described in Section 7.1.2. Reformatted from [44]. . . . . . . . . . . 146 7.2 Effect of the design intent parameters on the overall designed structure and assembly process in the formal ABM described in Section 7.2.2. . . . . . . . . . . . . . . . . . . . . . . . 151 xxiii Abstract Collective robotic construction (CRC) is a newly defined field that addresses questions of the application of multiple robot systems (MRS) in the Architecture, Engineering, and Construction (AEC) industry. In comparison to existing practices in construction automation, which often deploy individual static industrial machines to assemble pre-planned building elements, CRC systems generally involve the development of custom machines that can fit within a suitcase, are mobile within their construction environments and can collaborate to assemble architectural structures that emerge during their assembly. This new field introduces a series of research questions, specifically related to the transformation of current practices and methods in the fields of AEC, as CRC systems have the potential to disrupt the entire life cycle of a building, from design to disassembly. Although research in the field of CRC is currently expanding, it tends to be highly discipline-specific. This often results in the development of either generalized methods that overlook major constraints from other disciplines or highly customized approaches, which are applicable only to the specific CRC system under consideration. As a result, fundamental research questions regarding system integration for real architectural assembly and the overall co-design of the CRC system remain unanswered. Therefore, the major objective of this doctoral research is to illustrate how co-design, an approach to design that engages relevant disciplines in concurrent and feedback-driven xxv Abstract development, is facilitated by integrative computational design. This approach can effectively address the four key themes of CRC - construction materials, robotic systems, assembly algorithms, and architectural design - resulting in the development of architecturally relevant CRC systems. This is achieved through the presentation of a novel CRC system that utilizes timber, a building material found in the AEC industry. In the co-design process, this dissertation places particular emphasis on architectural design, as a primary theme that is often overlooked in research on CRC. It is explained that the comprehension, development, and analysis of methods for architectural design in CRC are important measures for realizing the full potential of such a novel approach to construction automation. Various approaches to architectural design are conceptually defined and formalized using the presented CRC system in order to demonstrate the range of approaches that can be taken in CRC. As such, this dissertation contributes to shaping a construction paradigm where real architectural structures can be robotically assembled using teams of small, mobile machines. xxvi Zusammenfassung Collective robotic construction (CRC) ist ein neu definiertes Forschungsfeld, das sich mit Fragen zur Anwendung von Multi-Roboter-Systemen (MRS) in der Architektur-, Ingenieur- und Bauindustrie (AEC) beschäftigt. Im Vergleich zu bestehenden Praktiken der Bauautomatisierung, bei denen häufig einzelne, stationäre Industriemaschinen zum Zusammenbau vorgeplanter Bauelemente eingesetzt werden, umfasst CRC in der Regel die Entwicklung anwendungsspezifischer, kompakter und möglichst einfacher Maschinen. Im Kontext von AEC kann die Zusammenarbeit mehrerer solcher Maschinen für die autonome Montage architektonischer Strukturen genutzt werden. Dieses neue Forschungsfeld wirft eine Reihe neuer wissenschaftlicher Fragestellungen auf, insbesondere in Bezug auf die Transformation aktueller Praktiken und Methoden in der AEC Industrie, da CRC-Systeme das Potenzial haben, den gesamten Lebenszyklus eines Gebäudes – von der Planung bis zum Rückbau – zu verändern. Obwohl die Forschung im Bereich CRC derzeit zunimmt, ist sie häufig stark disziplinspezifisch. Dies führt oft zur Entwicklung verallgemeinerter Methoden, die wesentliche Einschränkungen anderer Disziplinen außer Acht lassen, oder zu hochgradig angepassten Ansätzen, die nur für das jeweilige CRC-System anwendbar sind. Daraus ergeben sich grundlegende, bisher unbeantwortete Forschungsfragen zur Systemintegration für reale architektonische Montageprozesse sowie zur allgemeinen xxvii Zusammenfassung Co-Design-Methodik von CRC-Systemen. Das Hauptziel dieser Doktorarbeit ist es, zu zeigen, wie Co-Design, ein Ansatz, der relevante Disziplinen in einen gleichzeitigen und feedbackgesteuerten Entwicklungsprozess einbindet, durch integratives computergestütztes Entwerfen unterstützt wird. Dieser Ansatz kann die vier zentralen Themen von CRC - Baumaterialien, Robotersysteme, Montagealgorithmen und architektonischer Entwurf - wirksam adressieren und zur Entwicklung architektonisch relevanter CRC-Systeme führen. Dies wird anhand eines neuartigen CRC-Systems demonstriert, welches die robotische Fertigung von Holztragwerken ermöglicht und somit einen in der AEC-Industrie weit verbreiteten Werkstoff einsetzt. Im Co-Design-Prozess legt diese Dissertation besonderes Augenmerk auf den architektonischen Entwurf, welcher in der Forschung zu CRC oft vernachlässigt wird. Konkret wird hervorgehoben, dass das Verständnis, die Entwicklung und die Analyse architektonischer Entwurfsmethoden im Kontext von CRC wesentliche Schritte sind, um das volle Potenzial eines derart neuartigen Ansatzes zur Bauautomatisierung auszuschöpfen. Verschiedene Ansätze des architektonischen Entwerfens werden konzeptionell definiert und mithilfe des vorgestellten CRC-Systems formalisiert, um die Bandbreite möglicher Herangehensweisen im CRC zu veranschaulichen. Auf diese Weise trägt diese Dissertation dazu bei, ein Bauparadigma zu gestalten, in dem reale architektonische Strukturen anhand von Teams bestehend aus kleinen, mobilen Maschinen robotisch montiert werden können. xxviii Figure 1.1: Collective robotic construction (CRC) system composed of timber struts and robotic actuators that are required to work together in order to assemble timber structures. 1 Introduction Robots are increasingly prevalent in our everyday routines. What started with simple robotic toys for recreation and vacuuming robots for tidying homes is now progressing to more sophisticated robots that can navigate complex scenarios, like self-driving automobiles and humanoid robots offering assistance with household chores. Beyond robots entering our private lives in these and other ways, robots are also being incorporated into many professions, thereby catalyzing significant changes in various industries. Some examples of this include robots for assisting or fully automating surgeries in the healthcare industry, robots in factories for producing serial products or managing organizational logistics in the manufacturing industry, and robots for monitoring and harvesting crops in the agricultural industry. One additional sector that has seen a rise in robotics and more generally automation over the last several decades is the Architecture, Engineering and Construction (AEC) industry. Construction automation has gained momentum due to a variety of societal, ecological, and environmental challenges in the industry, among them are the low productivity of the industry, the high material and energy consumption during construction and often unsafe or improper working conditions [32]. The use of robots for construction automation is one 3 1 Introduction proposed solution to address these challenges. However, compared to other industries, the automation of construction entails additional constraints, making the introduction of robotic technologies a highly complex endeavor. This includes the geometric variation between different projects, the unstructured nature of most construction sites, the longevity requirements of buildings and the overall large-scale nature of the projects. As such, researchers and practitioners in the industry have been developing various approaches to robotic automation in order to determine the best way to tackle these additional challenges. Some of the various parameters that define the different approaches include the type, size, and number of robots, the materials and structural systems, and the phase of construction at which the robots intervene. Some specific examples of recent endeavors in the automation of construction using robots include on-site robotic factories which build entire buildings floor by floor [68; 23; 30], tower cranes for the automation of various parts of the building assembly process [22; 72], the re-purposing of industrial robots for the assembly or prefabrication of bricks, timber elements and other building elements made from other materials [12; 3; 33; 50], and a team of multiple mobile robots that assemble brick-like materials using principles inspired from termite mound construction [57] as seen in Figure 1.2. The last of these is an example of one emerging approach to integrating robotics into construction, known as collective robotic construction (CRC). CRC is a subset of construction automation that utilizes small, agile mobile robots working in teams to execute construction tasks [56]. Due to its reliance on the collective action of robots, CRC requires a fundamental rethinking of the role of robots in construction, in contrast to many other approaches that rely on single large machines for robotic automation. Furthermore, because the robots in CRC are no longer larger than the buildings they produce and have limited payloads, CRC demands that building systems be reconsidered. Another major factor to consider in CRC is what the robots do in the construction process and how this is coordinated, considering that the robots in a CRC system can do 4 1.1 Research Aim Figure 1.2: Various approaches to construction automation using different scales of robots including a large-scale building construction factory (Top Left) [68], a tower crane (Top Right) [72], an industrial robotic arm on mobile platform (Bottom Left) [12], and a team of bespoke mobile machine (Bottom Right)[57] more than just erect the structure. To address these questions, research on CRC necessitates the merging of research from the fields of distributed computing, robotics, and architecture, presenting an excellent opportunity for co-design, where interdisciplinary teams must collaborate to develop innovative systems. 1.1 Research Aim The aim of this dissertation is to introduce co-design methods for the development of novel CRC systems and thereby showcase the transformative effect that such an approach to automation in construction can have on the AEC industry. The primary objective is to demonstrate how interdisciplinary collaboration, facilitated by integrative computational design, can effectively address the four key themes of CRC — construction materials, robotic systems, assembly algorithms, and architectural design — ultimately leading to the development of architecturally relevant CRC systems. This is achieved through 5 1 Introduction the presentation of a novel CRC system for the assembly of timber structures. This dissertation places particular emphasis on illustrating the significance of integrating architectural design as a primary theme within the co-design framework. It is explained that the comprehension, development, and analysis of methods for architectural design in CRC are essential elements for realizing the full potential of this novel approach to construction automation. This dissertation was completed in the context of Research Project 19 (RP19) in the Cluster of Excellence Integrative Computational Design and Construction for Architecture (IntCDC at the University of Stuttgart. The IntCDC serves not only as a platform for the interdisciplinary investigation conducted but also as a backdrop for the comparison of various approaches to the robotic automation of construction. 1.2 Dissertation Structure This is a cumulative dissertation composed of five scientific peer-reviewed publications, Article A to Article E [38; 44; 36; 39; 37]. The broader context of the dissertation is presented in Chapter 2, which includes research on multiple mobile robotics and current practices of computational design and digital fabrication in architecture. Chapter 3 describes the two major research objectives and related research questions of this dissertation. Chapter 4 continues with the state of the art in which the literature review conducted in Article A is summarized in Section 4.1.1, giving further contextualization to the research. Chapter 5 summarizes the methods developed in the remaining publications, Article B to Article E. This is followed by Chapter 6, which has sections dedicated to the five scientific peer-reviewed publications. The next chapter, Chapter 7, presents the results based on the two major physical experiments conducted in the research. Chapter 8 discusses the results in relation to the research questions. Finally, Chapter 9 provides an overall conclusion, giving a broader perspective and potential future directions of the research. 6 1.2 Dissertation Structure At the end of the dissertation, there are four appendices that contain further relevant information for the dissertation. Appendix A contains further peer-reviewed publications related to the dissertation, in which the author made considerable contributions. Appendix B included information on additional educational workshops relevant to the main objectives of this dissertation that were organized and led by the author. The workshops provided a testing ground for the methods developed in this dissertation as well as a platform for educating architectural students on CRC. Appendix C is dedicated to master’s thesis projects that the author supervised. The projects provide further insight into potential future directions of the research. Appendix D is a list of project funds that contributed to this dissertation, detailing the associated researchers and principal investigators on the projects. 7 Figure 2.1: Initial physical prototype of the robotic actuator for the CRC system discussed in this dissertation. 2 Context and Background 2.1 Multiple Robotic Systems Since the introduction of multiple robot systems or multi-robot systems (MRS) in the 1980s, there has been a growing interest in their study and development. MRS are groups of mobile robots that cooperate within the same environment to achieve complex tasks [14]. MRS operate fundamentally differently from single-robot systems that work in isolation in their individual environments. They rely on the coordination and collaboration of numerous machines in order to accomplish any given task. Parker [55, p. 921] summarized the motivation behind research on MRS over single robot systems with the following arguments collected from other researchers [6; 1]: • The task complexity is too high for a single robot to accomplish. • The task is inherently distributed. • Building several resource-bounded robots is much easier than having a single powerful robot. • Multiple robots can solve problems more efficiently by utilizing parallelism. 9 2 Context and Background • The introduction of multiple robots increases robustness through redundancy. In the early stages of MRS development, researchers integrated knowledge from various domains, including distributed computing, biology, and autonomous robotics, to create simulations of or physical MRS capable of accomplishing basic tasks like traffic management, box manipulation, and foraging [6]. This required rethinking various aspects of robotic systems, including communication, task and motion planning, robot mechatronic design, localization, and coordination, as well as considering the transition from a single robot to multiple robots. With these initial successes and continued technological advancements in both hardware and software, MRS are currently being developed into highly robust robotic systems. As a result questions regarding their practical applications, beyond the simple tasks already mentioned, are becoming more important. Research on MRS is therefore shifting to investigate real-world applications, marking an important transition in the field. Nonetheless, there are only a limited number of real-world cases illustrating the integration of MRS into our daily lives, which contrasts the earlier discussion of robots being integrated into both the public and private spheres. This is primarily due to the heightened complexity involved when machines cooperate, which is compounded by the relative newness of the research field. One notable exception is the use of mobile robotic systems in warehouse management, exemplified by Amazon Robotics (formerly Kiva Systems) [10]. In such systems, mobile robots autonomously navigate around warehouses to replace conveyor belt systems for the distribution of goods. The general lack of MRS in the real-world confirms the fact that research is still required on MRS to make them robust enough for such applications. One application of MRS as introduced in Chapter 1 is construction, also known as CRC [56]. CRC proposes a significant departure from other forms of automation in construction. The machines are not fixed in space, are generally smaller in size than the structures they work on, have lower energy 10 2.2 Computational Design and Digital Fabrication requirements, and are typically based on custom designs tailored to the building systems they assemble. Researchers from various disciplines have been investigating CRC, as there is further potential for construction beyond the cost, time, and robustness benefits of MRS on their own. The additional benefits of working with multiple mobile robots specifically for construction are [44]: • the ability of the robots to work around the clock; • the ability of the robots to freely navigate; • the ability to the robots to actively respond to site, changing architectural design, or other constraints; and • the ability of the robots to inhabit the structure gives them the potential to do more construction activities than just assembly. Although it overlaps with research on MRS, CRC introduces many new research questions relating to topics such as architectural design. Understanding the development of CRC systems, the methods, necessary for their development and the capabilities of robots in construction processes are crucial questions in CRC research 2.2 Computational Design and Digital Fabrication Advancements in technology, both in computational design and digital fabrication, are transforming the field of architecture [8]. Although investigations with computers started in the 1960s, architects only began to extensively incorporate computers to aid in the design of buildings in the late 1980s and early 1990s, with the democratization of computer-aided design (CAD). Computers in the field of architecture initially served as tools for increased efficiency and productivity, but they eventually evolved into instruments for exploring innovative architectural forms that were challenging or even impossible to envision through traditional architectural handcraft. More recently, the computer is no longer thought of as a tool for design but 11 2 Context and Background as part of a new design thinking paradigm called computational design. In computational design, computation serves not only as support for architects but also as a means for rethinking the design process altogether [48; 69]. Computational design is enabled by the evolution of methods and tools for CAD, which now allow for the simulation of the complexities of architecture, rather than just the mere modeling of it. Alongside the progress in computational design focused on building design and visualization, the discipline of architecture is also being expanded with new forms of digital fabrication, which are reshaping how buildings are fabricated and constructed. Similar to computational design, digital fabrication was initially employed in the AEC industry for economic reasons but quickly transitioned to encompass the exploration of new types of structures and forms [13]. At the beginning, computer-aided manufacturing (CAM) machines were repurposed from the manufacturing industry for mass production. The benefits of such machinery were thereafter harnessed for the customization of building components. With the advent of technologies like autonomous robots and large-scale 3D printing, the range of machinery available for fabrication and assembly processes has expanded. This expansion is prompting architects to reconsider their design processes and, consequently, what can be achieved in architectural design and construction. 2.2.1 Co-Design in Architecture As computational design and digital fabrication evolve to address more complex challenges, the relationship between them is becoming more important. To understand what a specific digital fabrication process can create, computational methods are necessary for conceptualizing designs with it. Furthermore, with the additional complexities of both, expertise from outside the discipline of architecture is becoming necessary to aid in the creation of new methodologies that the technologies mandate. Consequently, architects are collaborating in interdisciplinary teams to enhance their individual expertise and reevaluate approaches to architectural design. Co-design is one approach 12 2.2 Computational Design and Digital Fabrication for integrating knowledge from various disciplines into a single framework. Co-design, which is a term that is utilized in varying contexts, has recently been defined as an approach that seeks to combine research on design, engineering, manufacturing, and construction to unlock the full potential of digital technologies in the context of AEC [32]. It aims to collectively rethink (i) planning and engineering methods, (ii) manufacturing and construction processes and (iii) material and building systems n the AEC industry, thereby breaking away from the incremental, discipline-based innovation currently prevalent in the field [32]. Achieving this goal involves employing an integrative computational methodology that negotiates the advantages and limitations from technology across the different disciplines participating in the development. As the guiding principle of the IntCDC, various examples of the implementation of co-design are showing its promise to make fundamental change in AEC and achieve the goal of creating "a high-quality, liveable, and sustainable built environment" [32]. This includes new methodological frameworks for the design and construction of segmented timber shell structures [71] and large-scale fiber composite building components [16]. 13 Figure 3.1: Final physical prototype of a robotic actuator with the screwdriving effector, utilized for connecting timber struts. 3 Research Objectives and Questions CRC is a radically different approach to construction automation, with the potential to revolutionize the AEC industry, as touched upon in the previous chapters. However, as the field is relatively new with reviews on the subject only being published in the last few years, numerous research questions still need to be addressed to make CRC a viable approach compared to current practices in construction automation. As such, this dissertation is organized based on two primary research objectives, each accompanied by a distinct set of research questions. 3.1 Co-design of CRC Systems Due to the multifaceted nature of CRC, which combines fields of computer science, robotics, and architecture, the development of CRC systems presents an ideal case for co-design. The compact and cooperative nature of the robotic agents in CRC requires a fundamental rethinking of construction automation, which means the collaborative conceptualization and development of novel CRC systems. Several critical interdependencies underscore the significance 15 3 Research Objectives and Questions of this collaboration. For instance, the limited payload capacity of the robots directly impacts the choice of building components and materials with which they can work. These chosen materials, in turn, establish a close relationship with the overall building system, which is further influenced by the kinematic reach of the robotic agents. Moreover, the coordination of robot kinematics in collaborative efforts prompts careful considerations regarding parallel assembly possibilities and the path planning necessary to accomplish it effectively. To comprehend and navigate these intricate relationships successfully, it is imperative to foster a robust interdisciplinary collaboration between architects and engineers. This collaborative synergy can be achieved through the application of computational co-design methodologies, which facilitate integrative research that spans across disciplines, leading to innovative CRC systems. As such, the following research questions arise: • How can the utilization of co-design principles facilitate the development of architecture-specific CRC systems? 3.2 Architectural Design in CRC The design of an architectural structure, tailored for the assembly or subsequent reconfiguration by a CRC system, constitutes an important aspect within the co-design of CRC systems. Architectural design in CRC is a highly intricate endeavor, necessitating the meticulous consideration of a myriad of parameters, including architectural application and the capabilities of the robotic system, such as its expansive workspace, design aesthetics, and potential fluctuations in the construction environment. As the consideration of the mobile robots should be included in early stages in the architectural design process, the overall process for architectural design in CRC often requires a complete re-imagining of established processes. Furthermore, the question of how the design is communicated to the robots and when the design is created relative to when it is constructed impacts the entire design-to-construction workflow in CRC. Beyond answering the question of what CRC systems can 16 3.2 Architectural Design in CRC build, there are many research questions relating to architectural design in CRC. The primary research questions related to architectural design in CRC discussed in this dissertation are: • What methods are currently utilized for the architectural design of structures with CRC systems? • How can computational methods be conceptualized and formalized for the architectural design of structures with CRC systems? • How can design intent from architects be represented in CRC workflows? • How do different architectural design methodologies and their corresponding workflows affect the phases of construction and interaction from architects in CRC processes? 17 Figure 4.1: Kinematic chain composed of two robotic actuators and one timber strut. 4 Current State of the Technology 4.1 Collective Robotic Construction CRC is a term that was defined in a publication from 2019 to describe state of the art research on the subject [57]. The authors define CRC as a subset of automation in construction that "focus[es] on multi-robot systems autonomously building structures far larger than individual robots" [57]. Other terms for research on this topic include swarm construction [67], multiple robot construction [2], and distributed robotic construction [47]. Self-organized construction is also a term that has been used to describe a similar but narrower in scope field of research. Self-organized construction is motivated by the coordination of robotic agents through local stimuli, therefore focusing on decentralization in the coordination of the robots [15]. Another major difference is the exclusion of robotic systems, which rely on their own embodiment to construct a structure. Conversely, CRC includes investigations on systems with varying degrees of centralization and on modular robotics in which the self-assembly of the robots is examined in the context of 19 4 Current State of the Technology construction. In research on CRC, there are four major themes of investigation: construction materials, robotic systems, assembly algorithms, and architectural design. The construction material theme is concerned with the selection and design of the material to be used, considering also how the material can be assembled together. Robotic systems encompass not only the examination of robotic mechatronic hardware but also topics of sensing and communication that relate to the entire robotic system. Assembly algorithms deal with the topic of coordination, involving sequencing the assembly process, task allocation, path finding and path planning. Architectural design refers to the process of deciding what the robots should assemble, which can involve investigations of structural analysis and life-cycle assessment. Through the systematic investigation of these themes, there have been various successful demonstrations of CRC. The range of which is explained through the following examples, sorted by year: • wheeled robots for the assembly of stacked plastic bricks inspired by logic from termites [57; 73]; • aerial construction of foam bricks into towers by quadrocopters [4]; • a custom swarm construction system composed of custom robots called Fiberbots for the manufacturing of tube structures out of fiber composites [26; 27]; • heterogeneous robot teams that are designed specifically for the assembly of filament structures [76]; and • a framework for aerial additive manufacturing composed of drones that can be used for simultaneous 3D printing and monitoring of the printed structure [79]. These are a few of the research projects that are part of the larger field of research, which contains over 100 publications, as found in Article A. The majority of research on CRC is "limited to work with relatively small 20 4.1 Collective Robotic Construction assemblies, with few robots, in controlled [or simulation] environments, and with the use of simulant materials" [57]. As such, this dissertation employs co-design to create a CRC system that can create architecturally relevant large assemblies out of an existing building material in the construction industry. To the author’s knowledge, there are only a few projects that work with real building materials. Most projects use highly mechanical, non-structural materials that simulate those for real construction. The materials utilized range from foam bricks [63], plastic voxel materials [24], fibrous materials that represent structural fiber materials [77], plastic truss elements [53] to toothpicks [51]. Two of the above mentioned CRC systems use real building materials. The research from Kayser et al. deals with fiberglass composites, which, although available as a building material, are not a prevalent choice within the construction industry [26]. Zhang et al.’s 3D printing uses cementitious material, which is more common in the industry but typically requires integration with other construction materials or systems to achieve large-scale applications [79]. Furthermore, in both [26] and [79], the individual robots do not directly collaborate but rather work in parallel. This dissertation makes a contribution by presenting a CRC system in which robots must collaborate, which brings significant additional challenges to all areas of development in CRC. 4.1.1 Existing Research on Architectural Design in CRC Although architectural design is a major theme in the development of CRC systems, it is often overlooked. This is due to the fact that a large portion of research on CRC stems from the field of MRS, which is mostly based in engineering contexts. Yet, architectural design is one of the main research themes in CRC. Research on architectural design extends beyond simply defining what architectural structures are feasible; it also plays a pivotal role in determining how architects can effectively engage with these technologies, ultimately making them more accessible to a broader range of professionals. 21 4 Current State of the Technology The application of such construction automation systems will remain difficult if architects cannot design and work with CRC systems. Recognizing this, the major contribution of Article A is a review of the state of the art on the theme of architectural design in CRC. Article A addresses the first research question under the theme of architectural design in CRC, listed in Chapter 3. More specifically, Article A: • develops and explains a categorization system of CRC based on approaches to architectural design; • analyzes 122 peer-reviewed publications collected through an online database; and • discusses trends and further research directions for architectural design in CRC. A concise overview of the categorization system is provided below. The categorization system presented for classifying approaches to architectural design in CRC is split into three dimensions: design description, goal specification, and execution [38]. The first dimension focuses on whether the architectural design is known before construction begins. Approaches are either pre-defined designs, in which everything is known before construction, or emergent designs, in which the outcome is not explicit till the end of the construction process. The goal specification dimension defines how architectural designs are communicated to robots. This can be with exact blueprints in pre-defined approaches or as user-defined conditions and functional goals in the case of emergent approaches. The execution level describes when the robotic planning of the associated architectural design occurs, with two types: real-time execution, where architects make decisions on the fly, and simulation execution, where fine-tuning and design iteration can occur. The last dimension is crucial for understanding the entire design process, as architectural design in CRC is linked to the execution of robotic actions and material deposition by mobile robots. This categorization system 22 4.1 Collective Robotic Construction can be used to understand how architectural design is approached in research on CRC [38]. A detailed description of the categories, using examples from 122 collected publications, can be found in Article A. 4.1.1.1 ABMS One promising direction for the development of methods for architectural design in CRC, noted in Article A is agent-based modeling and simulation (ABMS), an approach of behavioral artificial intelligence (AI). ABMS is a method for the investigation of complex adaptive systems. Also known as individual-based modeling, ABMS involves the modeling and simulation of autonomous decision-making agents [46]. The process of developing an agent-based model (ABM) involves defining and refining the agents’ behaviors or simple rules, which dictate how they act and interact within a shared environment. The goal of ABMS is to obtain insights into collective patterns, structures, and behaviors of the agents, which is often achieved through iterative computation. Due to its focus on individual entities, ABMS has gained popularity because it enables the exploration of systems that are otherwise difficult to predict or understand [19]. Thus, although ABMS originated in the field of computer science, it has now expanded into other fields, including biology, economics, epidemiology, engineering, and the social sciences. Among the fields that have seen an increase in the utilization of ABMS are those related to AEC [29]. This is because architecture can be understood as a complex system based on the interactions of varying individual entities, such as people or stakeholders, discipline domain knowledge, building elements, technical systems, and digital or physical builders [66]. Additionally, ABMS possesses several attributes that render it well-suited for application in the AEC domain. Schwinn concisely outlines these qualities [61], noting that ABMS: • exhibits a foundation in bottom-up principles, enabling both exploratory and goal-oriented investigations; 23 4 Current State of the Technology • inherently relies on feedback mechanisms; • permits the exploration of diverse simulation scenarios with different agent types within a unified modeling framework; and • offers possibilities for enhanced communication to stakeholders and greater flexibility when compared to similar computational methods. In research in the realm of AEC, the versatile use of agent-based models (ABMs) extends across all phases of construction, including design, engineering, planning, construction, operation, and eventual demolition. Some examples include the modeling of crowd behavior for architectural design optimization based on visitor comfort [62], a multi-agent system that helps mediate information from distributed resources for generative design [11], and balancing comfort and energy consumption during the operational phase of a building [31]. Among the growing number of examples of ABMs in architecture, there are some ABMs that have been created for architectural design exploration within CRC. Two recent examples are Pietri and Erioli’s ABM for the design of fibrous structures assembled by drones [58] and Kayser et al.’s ABM for the design of tubular structures with the Fiberbots [26]. Both instances are based on the boid ("bird-oid") model, originally proposed by Reynolds in 1987 to understand flocks, herds, and schools [59]. The fundamental setup of the boid model involves locomotive agents guided by three simple behaviors: separation, alignment, and cohesion. Pietri and Erioli, along with Kayser et al., built upon this model by incorporating parameters from their respective CRC systems. This manner of adjusting the boid model is a common practice in the realm of ABM experimentation for architectural design, both within and outside of the CRC context. This includes the earliest examples of architectural design exploration with ABMS [49] to the more recent sophistication of ABMs to realize large-scale built projects [71]. However, developing an ABM for CRC based on the boid model comes with limitations. In both known cases, the trails of agents or locations where 24 4.1 Collective Robotic Construction the agents travel in the simulation are taken as the paths that the robots will eventually travel in the real-world. This has two major limitations. First, following complex 3D curves that result from boid models in the real-world is not achievable with all of the types of robots used in CRC systems. One example is the robotic system presented in this dissertation. Second, when the ABM is run entirely in simulation and subsequently post-processed for execution, the assembly process loses its adaptability and responsiveness to environmental, design, or other constraints. This dissertation aims to show how ABM can be used in CRC beyond adaptations of the boid model and give specific insights into the formalization of different conceptualizations of ABMs, which can be utilized in simulation and during the assembly process. 25 Figure 5.1: Four robotic actuators in two kinematic chains ready to start the process of assembling a large-scale in-plane assembly. 5 Research Structure and Methods This dissertation discusses the development of a novel CRC system for timber construction. It builds upon preliminary research from the author [43] by introducing co-design as a guiding principle for its further development. Beyond the previously discussed review in Article A, this dissertation is composed of four additional publications, Article B to Article E, which describe the pursuit of the research objectives outlined in Chapter 3. Descriptions of the overall CRC system and its co-design are discussed in Article B and Article C, addressing the first research objective. The research was initiated through the presentation of a CRC system that leverages building material as part of the robotic kinematic system for collective construction of in-plane timber structures, as reported in Article B and showcased with three short-horizon assembly tasks. At the system level, the process of co-design continued in Article C, in which the CRC system was further developed for the assembly of large-scale in-plane structures that require longer sequences of robotic action. Three publications, Article B, Article D, and Article E, tackle the second 27 5 Research Structure and Methods research objective related to architectural design in CRC, approaching it from various angles. Article E introduces ABMS methods for designing architectural structures with MRS. It describes two approaches on a conceptual level as well as their formalization using the developed CRC system. It also discusses how each approach situates architectural design in an overall construction workflow in CRC. Article B provides a more detailed description of one of the methods presented in Article E and describes its interaction with the custom task and motion planner that is based on Logic-Geometric Programming (LGP) [70]. Article D builds upon one of the approaches in Article E and extends it by including the full robotic planning of the structure into the ABM. In addition, this paper places a strong emphasis on interactivity, showing the levels of control an architect can have in design and assembly processes with CRC. The approach in Article D was developed and implemented in the context of the in-plane prototype assembled with the developed CRC system, described in Article C. The following sections provide a concise overview of the developed CRC system and the architectural design methods presented in the publications comprising this dissertation. The publications can be found in Chapter 6. 5.1 Modular CRC System for Timber Construction The co-design of the developed CRC system involved explorations of the four major research themes in CRC: robotic systems, construction materials, assembly algorithms, and architectural design. The author was responsible for the development and realization of all major principles of the system, working closely with researchers in the domains of mechatronics and task and motion planning (TAMP). The investigations of architectural design were done solely by the author. 5.1.1 Concept Overview The developed system combines two concepts for robotic development seen in research on CRC: one where the robot functions as a manipulator of 28 5.1 Modular CRC System for Timber Construction construction materials, and another where the robot assumes the role of the construction material itself [44]. To clarify this, an example of the robot as a manipulator is modular robotic modules that through active connection mechanisms can self-assemble into furniture scale objects [65] and an example of the robot as a manipulator, which is the more common type of CRC system, is ground robots that assemble ramps using foam blocks and compliant sandbags [60]. The developed system is composed of two elements: timber struts and robotic actuators. It draws inspiration from previous explorations of truss climbing robots [53; 78], but differs in that the robotic actuators are completely unable to move or act independently. The robotic actuators must leverage the timber struts in order to allow themselves to move [44]. The timber struts, therefore, serve not only as building material but also as part of the robot’s body, forming kinematic chains that enable movement within the system. The CRC system relies on both robotic actuators and timber struts to work together and in parallel to assemble timber structures. 5.1.1.1 Design Space Due to their modular nature, kinematic chains or a series of linked robotic actuators and timber struts can break apart or rearrange to form kinematic chains with varying degrees of freedom (DOF). In an idealized version of the developed system, robotic actuators and timber struts can be assembled in any arrangement with the goal of obtaining 6 DOF to allow kinematic chains to reach any point at any orientation in 3D space. The typology of a kinematic chain has a direct influence on its ability to perform assembly-related tasks [44]. Although generally more robotic actuators equate to larger kinematic freedom of a kinematic chain, the relationship is not direct, as certain typologies restrict or duplicate additional DOF that are provided by additional robotic actuators. While having extra robotic actuators does provide more rotational possibilities in kinematic chains, it typically leads to increased complexity in the mechatronic design and control of the robotic system. When dealing 29 5 Research Structure and Methods with a compact mobile machine, it becomes challenging to provide adequate support for longer cantilevers and heavier loads in kinematic chains featuring additional robotic actuators and timber struts. Therefore, a study of kinematic chains of varying typologies was conducted to determine the minimal design necessary for assembling timber struts in Article B. The study was limited to typologies in which robotic actuators only attach to the timber struts along their length, not at the ends, resulting in four out of six orientations around the timber strut. This is due to the complexity that arises mechanically when robotic actuators need to be attached to both sides and ends of a timber strut. The study of the kinematic chains was based on the analysis of two tasks required for the assembly of structures: • Locomotion explains where in 3D space a kinematic chain can move on its own. • Manipulation describes how the kinematic chain can move an additional timber strut when it is gripped at its end. Through this study, it was determined that kinematic chains composed of two robotic and one timber strut can collaborate to build in-plane timber structures [44]. Figure 5.2 shows a kinematic representation of such kinematic chains. This requires the kinematic chains to increase their complexity by including an additional timber strut at specific moments in the construction process. With the initial goal of showcasing the system’s ability to construct large-scale structures, the decision was made to first demonstrate the developed system using this minimal typology of kinematic chains [44]. 5.1.1.2 Robotic System The main element of the robotic system is the robotic actuator (Figure 2.1 and Figure 3.1) [36; 44]. The major features of the robotic actuator are a rotational joint and two opposite-facing grippers. The robotic actuators can attach to timber struts by opening and closing the grippers on either side of the rotation joint. In addition to the ability of each gripper to attach to or release from the 30 5.1 Modular CRC System for Timber Construction Figure 5.2: Kinematic representation of a kinematic chain with two robotic actuators and one timber strut [36]. timber struts, they can slightly lift them. This ability of the grippers stems from the consideration of the assembly process. It allows for gripped timber struts to move from the exact plane from which they were picked up to avoid any potential contact and thus friction with anything directly below them. There were two additional requirements from the perspective of architectural design that informed the design of the robotic actuator. First, the rotational axis should be unlimited, allowing the robotic actuator to rotate to any angle and giving rotational freedom to timber structures that could be assembled. Second, the region required for opening and closing the gripper should be minimized. This allows timber struts to be assembled as closely together as possible. Beyond the robotic actuator, another important feature of the robotic system is the mobile timber connection mechanism, or effector, which allows for the fixation of timber elements [36]. Three major criteria were utilized in order to decide what connection strategy to utilize in the system. The criteria, as discussed in Article C, are: • The connection strategy should be based on methods that are already 31 5 Research Structure and Methods used in the timber construction industry. • The connection should be completed with as little force as possible. • The connection should be reversible to allow timber struts to serve as both part of the structure and part of the robot body at different moments in the assembly process. Using these defined criteria, screwed connections were selected for further development. The other connection types considered were milled wood joints, nails, glues, dowels, and bolts. The physical embodiment of the screwing mechanism is based on a drill with an automatic screw feeder (Figure 5.3)[36]. It sits on half of the robotic actuator, allowing the actuator to continuously rotate. The effector contains two major mechanisms: one mechanism for screwing two screws simultaneously into timber struts and another mechanism for feeding screws through the effector using plastic belts. The decision to design the effector to drill two screws simultaneously was to avoid any pivotal rotation that could occur through the insertion of only a single screw. The final element of the robotic system is an external motion capture system [44]. The motion capture system is used to provide position and orientation estimation for both robotic actuators and timber struts in all the physical experiments conducted with the developed CRC system. 5.1.1.3 Material System The material system is composed of timber struts with a square cross section of 50 mm by 50 mm [44]. The length of the timber struts is a variable in the formation of kinematic chains and in the creation of timber structures. The chosen lengths are reported on in the respective publications on the co-design of the system, Article B and Article C. The timber struts are made out of spruce, a common softwood used in timber construction. Grooves were milled into the timber strut along two sides where the gripper paddles of the robotic actuator make contact. This co-design between the robotic actuator 32 5.1 Modular CRC System for Timber Construction Figure 5.3: Effector for assembling timber struts together (A), including diagrammatic representations of how the two major mechanisms for screwdriving and screw feeding function (B-C) [36]. and building material enables more stable connections, which are crucial for preventing unexpected movement or errors in the system. Another key factor of the material system is that timber struts are connected via the screwdriving effector along the long face. Timber struts cannot be connected end-to-end or side-to-end, which must be considered in the design of timber structures to be assembled by the system. 5.1.1.4 Robotic Basic Motion Primitives & Assembly Tasks Key to the development of the CRC system was also the consideration of how robotic basic motion primitives could be translated into tasks required to assemble in-plane structures using the developed CRC system. The basic motion primitives are as follows [36]: • Open Gripper: This motion primitive activates the gripper to allow the gripper paddles to spread apart. This action allows the robotic actuator 33 5 Research Structure and Methods to release from a timber strut. Once this motion is executed, the gripper paddles are behind the contact surface of the robotic actuator that touches the timber strut. This full retraction avoids any potential collisions as the robotic actuator moves over timber struts. The lifting mechanism in the gripper slightly lowers the gripped timber strut when it opens. • Close Gripper: This motion primitive is the opposite of the previous primitive, dealing with the closing of the gripper paddles around the timber strut. In contrast to opening, closing the gripper slightly lifts the timber strut. • Joint Rotation: This motion primitive involves rotating the main axis of the robotic actuator. • Drill Screws: This motion primitive involves drilling the current set of screws in the screwdriving effector and then returning the drill bits to the top of the effector. • Feed Screws: This motion actuates the mechanism for screw feeding in the screwdriving effector. Two other primitives were added with the enhanced co-design of the robotic actuator described in Article C, following the introduction of a new mechanism featuring a linear actuator. Both relate to the realignment of the robotic actuators in a kinematic chain [36]. These additional two are: • Actuate Tilt Adjustment: This motion primitive is accomplished by extending the linear actuator to engage the tilt adjustment mechanism, causing the robotic actuator at the opposite end of the kinematic chain to be lifted. • Release Tilt Adjustment: This motion primitive is the opposite of the previous. Through the combination of these basic motion primitives, three assembly tasks were designed for assembling in-plane structures [44]. These tasks also 34 5.1 Modular CRC System for Timber Construction formed the basis for developing the custom TAMP planner [44]. The tasks assume a maximum configuration of two robotic actuators and two timber struts in each kinematic chain. For all tasks, a major assumption is that the kinematic chains performing the assembly tasks are set upon a layer of timber struts. The kinematic chains operate above this layer, while new timber struts are assembled within it. The three assembly tasks are [44]: • Locomotion is the action of bringing a kinematic chain from one location to another. • Dynamic kinematic chaining is the action of changing the typology of a kinematic chain. • Transportation is the action of moving a timber strut into place for assembly into the structure. Each concerns a different number of kinematic chains and robotic actuators. Locomotion only concerns a single kinematic chain. Dynamic kinematic chaining requires an additional robotic actuator to facilitate the rearrangement of elements. Transportation involves two kinematic chains that must collaborate to accomplish this task. Further explanation of these can be found in Article B. Although dynamic kinematic chaining is not always necessary for the assembly of structures, it can be utilized for rearranging robotic actuators within the system and for adjusting the lengths between actuators in kinematic chains. This can be helpful in planning the assembly of structures and can impact the overall design space [44]. 5.1.1.5 Physical Experiments A key contribution of this dissertation lies in its validation of the proposed CRC system through a series of comprehensive physical experiments. Although simulated environments are often favored in CRC for their efficiency and cost-effectiveness, this research emphasizes the essential role of real-world validation, outlining several critical reasons for its necessity. First, despite the many advancements in simulation technology, the 35 5 Research Structure and Methods physical world remains remarkably complex and unpredictable. Simulating every aspect of the real-world accurately is an immense challenge. The CRC system operates in environments with a myriad of variables, such as tolerances in machined materials, unforeseen friction, and other obstacles. These are intricacies that are either difficult to replicate in a simulation or entirely impossible to anticipate. Therefore, relying solely on simulated experiments may lead to incomplete or inaccurate conclusions about the real-world performance of CRC systems. Second, physical experiments offer a unique opportunity to encounter and address challenges that are specific to real-world scenarios. In simulation, certain variables or interactions may be inadvertently overlooked, only revealing themselves in the physical world. By conducting comprehensive physical experiments, the identification and mitigation of real-world constraints and issues can be achieved. Furthermore, physical experiments provide valuable empirical data that can be compared to theoretical predictions and simulations. This comparison allows for a more comprehensive assessment of the CRC system’s performance and provides an opportunity to refine the system based on empirical findings. The ability to validate the functionality of the CRC system through physical experiments strengthens the credibility of the research and bolsters confidence in the proposed construction methodology. The physical experiments were split into two parts: (i) a set of demonstrations to showcase the ability of the CRC to perform the robotic tasks required for assembly with the system [44], and (ii) the assembly of a large-scale in-plane timber structure [36]. The discussion of both follows in Section 7.1 and Section 7.2, respectively. 5.1.1.6 Digital Twin/Interface The computational backbone for executing the physical experiments with the developed CRC is a digital twin, as introduced in Article B. It serves as the means for the interdisciplinary exchange required for the assembly of timber 36 5.2 Agent-Based Architectural Design Methods structures. Specifically, the digital twin coordinates the design, planning, communication of instructions, and monitoring of the construction process [44]. It enables direct interaction with the architectural design methods developed and communication with software in which these methods can also be implemented. It can perform direct queries using the custom TAMP planner developed for the system. It can send instructions to the robot using serial communication and collect and manage data received from the external motion capture system, which is utilized for error correction and tolerance adjustment during the assembly process. The Unity Game Engine1 was utilized to develop a customized interface for the digital twin, enabling the system to operate either through teleoperation or autonomously [44]. 5.2 Agent-Based Architectural Design Methods This dissertation investigated ABMS for the design of structures assembled by MRS within the larger context of architectural design in CRC. Within this pursuit, the following contributions were made: • the presentation of conceptual models for developing ABMs for architectural design in CRC; • the formalization of ABMs in CRC that can be reproduced, reworked, or built upon, and that are implemented within software used in the architecture profession; • the validation of ABMs for CRC; and • the analysis of how different conceptualizations and implementations of ABMs can affect overall workflows, and thus phases of construction, in CRC. The first three are based on imperatives for the development of ABM, which are listed in many guides on ABMS [17; 45] and have been utilized 1 Unity Technologies: Unity, https://www.unity.com/ 37 5 Research Structure and Methods within architecture as well as in others disciplines, such as archaeology, as a means of advancing research in the field [7; 61]. The last is specific to understanding the impact of different designs on construction workflows in CRC and is specific to the context of architectural design. The discussion of all of the contributions begins in this section and then continues in Chapter 7. 5.2.1 Justification for ABM in CRC Beyond the advantages of ABMS in the general context of architectural design introduced in Section 4.1.1.1, there are several explicit benefits to utilizing ABMs in CRC [37]. Robotic construction by multiple mobile machines introduces a web of dynamic and interdependent interactions among the mobile machines, the material being assembled, and their environment. ABM offers a unique advantage by allowing for the modeling of these complex interactions. One of the key challenges in architectural design for CRC is understanding the opportunities and constraints of the physical CRC system [37]. ABMS allows for the embedding of the principles of the CRC system into the model. These principles can be derived from the constraints of architectural design, as well as from other interdisciplinary research areas embedded in the CRC. As a result, an ABM can provide a platform for simulating and testing architectural designs within both virtual and physical environments. Architects do not need full knowledge of the robotic system in order to create architectural designs that consider the capabilities of the robotic construction team. Another compelling reason to investigate ABM is its inherent flexibility. ABMs can adapt to a wide range of scenarios and construction contexts, making them a versatile tool. Whether the project involves drones or climbing machines on the robot side or 3D printing or timber building elements on the material side, an ABM can be tailored to represent the capabilities of the robotic agents involved and reflect the requirements of the material system. This adaptability allows architects to explore various construction methodologies and experiment with different combinations of robotic agents and material systems. 38 5.2 Agent-Based Architectural Design Methods The flexibility of ABMS also enables the development of models that serve various purposes throughout the overall construction process. By varying how ABM elements are defined, different design workflows can be explored [37]. This includes models that explore architectural design alone to models that can be used for both design and the execution of assembly processes with CRC systems. Finally, ABM offers a method for translating complex construction processes and design concepts into accessible, visual representations. The simulations generated with ABMs can be visualized to depict the interactions between multiple mobile robots as they assemble structures. These visualizations can serve as communication aids, allowing architects, engineers, and other stakeholders to gain a clear understanding of the construction process, anticipate challenges, make informed architectural design decisions in real-time, and drive other informed decisions. Moreover, the visual outputs of ABMs can facilitate collaborative discussions among diverse project teams and foster a shared vision for the architectural design. As CRC is reliant on interdisciplinary collaboration, the capacity of ABMS to aid in visualization and communication can be beneficial for ensuring success in CRC research and development. 5.2.2 Conceptual Models Three conceptual models for architectural design in CRC were developed in this dissertation. Two are explained in Article E, and the other one in Article D. Each of the approaches was defined using the essential questions for designing and implementing ABMs as defined by Macal and North [46]. The questions relate to defining the purpose of the model, as well as its main constructs, which include the definition of the agent, environment, behaviors, and agent system. The following subsections provide a brief overview of the approaches that have been developed. A summary is presented in Table 5.1. One of the approaches from Article E is discussed together with the one from Article D, 39 5 Research Structure and Methods Agent Construct Approach: Agent represents Building Material [44] Approach: Agent represents a Mobile Robot [44] Approach: Agent represents a Mobile Robot [39] Purpose Architectural Design Architectural Design & Robotic Path Planning Agent Building Material Mobile Robot Environment 2D or 3D Continuous Euclidean Space Behaviors Rules based on geometric and robotic constraints Rules based on robotic constraints Agent System Collection of Building Elements Team of Mobile Robots Table 5.1: Outline of major global design decisions for the each ABMS approach as they are very similar, except for the purpose of the model, which imposes more constraints on the definition of the main agent constructs. In detailing the models on a conceptual level, the aim is to ensure that each approach can be used by others in the future. 5.2.2.1 Approach: Agent represents Building Material Purpose of the Model. The purpose of the model is to derive architectural designs that can be assembled by a CRC. Agent. Although the most intuitive definition of an ABM for CRC is to 40 5.2 Agent-Based Architectural Design Methods define the agent as a mobile robot, it is also possible to define the agent as building material. Architectural structures emerge through the negotiation of the placement of the building material. This approach defines the agents by the geometric properties of the building material. Environment and Interaction Typology. The ABM negotiates the placement of the agents in continuous digital Euclidean space. Agents interact with others who are close to them in proximity. The environment should also include any information on the physical environment, which would geometrically affect the location of the agents. This can include, for example, obstacles where building materials could not be assembled. Behaviors. The behaviors or rules that govern the interaction of agents are controlled by geometrical and robotic constraints. Geometric constraints inform the position and orientation of the agents as a product of the material system. Robotic constraints ensure the feasibility of assembling the designed architectural structure, considering the kinematic reach of the physical robot. Agent System. The agent system is a collection of all of the agents, which together represent the structure that can be built by the CRC system. With this approach, the agent system is governed by a position-based approach over a force-based approach. This means that at each iteration, the position of the agents change as a result of a translation vector acting upon them. As the system is concerned with the location of building material, the velocity of the agents does not need to be modeled and therefore does not require a force-based approach. 5.2.2.2 Approach: Agent represents a Mobile Robot Purpose of the Model. When the agent represents a mobile robot, it is also possible to include the robotic path planning of the structures as a purpose of the model, alongside architectural design. The approach described in Article D includes an additional requirement for the model to output the robotic path planning [39], while the approach in Article E only considers architectural design as the purpose of the model. 41 5 Research Structure and Methods Agent. Regardless of the purpose for the model, the physical mobile robots of the CRC system are considered the agents. The agents are defined by the conditions or properties of the physical robots, including information on the robot’s physical geometry, current position, and orientation, as well as its kinematic reach. Environment and Interaction Typology. Similar to the approach where the agent represents building material, the environment is a continuous digital Euclidean space. It represents the area or volume in which mobile robots or agents can operate. It should include information that geometrically affects architectural design, but it also needs to contain information on the building material in the system. When planning is included in the purpose of the model, the environment should also include any information that would affect the planning of the mobile robots. Such information includes where material is fed to the robots when they are ready to assemble new pieces of building material into the structure. Behaviors. In this approach, the behaviors should represent how the robots act in the real-world. This can vary based on the robots being used. When the model isn’t used to build the structure, the behaviors can be high-level abstractions. However, in cases where planning is necessary, the behaviors should relate to exactly how the robots act in the real world, in order for the model to achieve its purpose. Too large abstractions of real robot behaviors lead to the generation of invalid path plans for the real robots. Geometric constraints associated with the material system can likewise be incorporated into the behaviors, especially in scenarios where these behaviors determine the final placement of building materials within the environment. Agent System. The agent system is a collection of all of the mobile robots. The state of the system as a whole does not represent the architectural design but rather a moment in the construction process. The status of the built structure is stored in the environment. Depending on the level of tuning or control required for planning, you can utilize either a position-based or force-based system. Force-based systems iterate the position and orientation 42 5.2 Agent-Based Architectural Design Methods of the agents based on their velocity. Therefore, using a force-based system can achieve greater control over robotic planning. 5.2.3 Software Implementation: ABxM These conceptual models were translated to formal models using a custom software framework for agent-based modeling, which was first introduced by Groenewolt et al. [18] and later released as ABxM [52]. The ABxM Framework is "... an open platform for experimentation with agent-based, aka individual-based, systems. The aim of the framework is to standardize research equipment, in this case the tools for modeling and simulation, in order to increase transparency of agent-based models, and repeatability of research results." [52] The reason for choosing ABxM is to achieve the goal of creating reproducible models using software that is widely accepted in the architectural community. ABxM is open-source, allowing for the recreation or further development of the specific implementations created for this dissertation. Furthermore, ABxM is one of the few toolkits for ABMS that was designed with architectural considerations in mind [61]. It is based on the software development kit (SDK) for the CAD software environment Rhino, making it accessible to practitioners in the architecture profession. Although it is based on the SDK for Rhino, ABxM can be utilized in other software such as Unity, AutoCAD2, and Revit3 using Rhino.Inside, which makes it accessible to an even wider audience. For developing the formal models using the ABxM framework, several software tools were utilized. Visual Studio4 was used to build on top of the core agent library, ABxM.Core. Testing of the implementation was conducted 2 Autodesk: AutoCAD, https://www.autodesk.com/products/autocad/overview 3 Autodesk: Autodesk Revit, https://www.autodesk.com/products/revit/overview 4 Microsoft: Visual Studio, https://visualstudio.microsoft.com/ 43 https://www.autodesk.com/products/autocad/overview https://www.autodesk.com/products/revit/overview https://visualstudio.microsoft.com/ 5 Research Structure and Methods in Rhino using Grasshopper, as well as in Unity. These software platforms allow for the visualization of models and interaction with them. Chapter 7 discusses the specific details of the formal models developed for this dissertation. The implementations discussed in Article B and Article E were made before the release of ABxM, while the later implementation discussed in Article D was made with the ABxM framework [52]. 44 Figure 6.1: Four robotic actuators in two kinematic chains in the process of assembling a large-scale in-plane assembly. 6 Publications 6.1 Article A: Architectural Design in Collective Robotic Construction 47 6 Publications Leder, S., & Menges, A. (2023). Architectural Design in Collective Robotic Construction. Automation in Construction, Vol. 156, No. 105082, pp. 1-15. DOI: 10.1016/j.autcon.2023.105082 This publication surveys and analyzes current implementations of CRC based on their approach to architectural design, as examined through a literature review. The publication synthesizes current architectural design approaches in CRC, aiming to understand their implications on the overall design and assembly process when considering CRC. The publication introduces a categorization system that allows for the classification of architectural design approaches in CRC based on three categories: design description, goal specification, and execution. Through the classification of 122 peer-reviewed publications, the main finding is the identification of current and future trends in architectural design for CRC systems, highlighting the critical role of design approaches in advancing CRC research. Generally, the publication guides architectural design for construction processes that utilize multiple mobile machines. It explains to architects how architectural structures built using CRC systems are currently designed and the implications thereof. This literature review in this publication was conducted by S. Leder. This includes the creation and execution of the research methodology, as well as all analysis of the results. The research was overseen by A. Menges. The manuscript preparation, including drafting of the text and figures, was conducted by S. Leder, with editorial revisions from A. Menges. S. Leder led the revision process, including preparing responses to all rounds of peer review, under the supervision of A. Menges. 48 Automation in Construction 156 (2023) 105082 Available online 9 September 2023 0926-5805/© 2023 Elsevier B.V. All rights reserved. Review Architectural design in collective robotic construction Samuel Leder a,b,*, Achim Menges a,b a Institute for Computational Design and Construction (ICD), University of Stuttgart, Keplerstrasse 11, 70174, Germany b Cluster of Excellence Integrative Computational Design and Construction for Architecture (IntCDC), University of Stuttgart, Germany A R T I C L E I N F O Keywords: Collective robotic construction(CRC) Swarm construction Multiple robot systems (MRS) Architecture Architectural design Construction automation Mobile robotics A B S T R A C T Building construction is one application of multiple robot systems (MRS) that is seeing a rise in research and is known as collective robotic construction (CRC). CRC specifically introduces a new set of research directions to those that already exist in the field of MRS, one of which is that of architectural design. As architectural design is essential to the process of construction, this paper presents a review of research related to architectural design in CRC. The reviewed literature is categorized based on three dimensions: design description, goal specification and execution. This categorization is utilized to detail approaches to architectural design and understand the im- plications of different approaches on the design process when considering CRC. By enabling architects with the ability to understand how to design architectural structures built by CRC systems and the implications thereof, this review aims to reduce the barrier for working with such novel construction automation systems. 1. Introduction Since the 1980s, advanced simulations from the field of distributed computing have been converging with autonomous robotic systems into a combined research field titled multiple robot systems (MRS). This emerging field involves the execution of high-level goals through the collaboration of multiple machines without fixed positions within the same environment. As initial studies are now being developed into highly robust robotic systems enabled by rapid technical developments on both hardware and software, questions of application are becoming more important for the future development of the field. Therefore, research is transitioning beyond the initial exploration of three simple tasks (traffic control, box-pushing, and foraging [1]) to complex real- world applications including clean up, search and rescue, security, and construction. Through these application-specific investigations, new research questions are being discovered and fields of research are being developed. Collective robotic construction (CRC) is one of such newly defined fields that addresses questions of MRS for application in construction [2]. CRC combines research from the three initial applications of MRS in order to deploy small scale mobile robots for the autonomous assembly of architectural structures that are larger than the machines that build them. This new field introduces questions related to architectural design, structural engineering and construction on top of the challenges of robotic systems already inherent in MRS. Although some challenges such as those of communication, localization and mapping can be directly adapted and implemented from already developed solutions from the field of MRS; Some topics like architectural design, assembly sequencing, material system and structural stability need to be newly researched for the successful development of CRC systems. From an engineering perspective, the development of CRC is highly quantitative in which the robotic system can be evaluated based on its overall effectiveness, robustness or reliability [3]. Therefore, CRC sys- tems are generally regarded as more time-efficient, less prone to extended failures, and scalable as compared to single robot systems. From an architectural perspective, on the other hand, questions of architectural design introduce more qualitative guidelines regarding the aesthetics or overall appeal of the resulting structure, such as its level of symmetry or the composition of space, in addition to quantitative guidelines related to topics like technical building requirements and life- cycle assessment goals. Due to this, architectural design as it relates to CRC can be highly complex in which architectural application, the capability of the robotic system, stability and design aesthetic are some of the many parameters that must be considered. Furthermore, when considering CRC as compared to existing ap- proaches to construction automation, including on- and off-site pre- fabrication factories and single-task on-site robots [4], the complexity of architectural design becomes even more clear. First, since the robots are mobile, the workspace of the robots in CRC is theoretically unlimited allowing for even larger architectural structures to be constructed. * Corresponding author. E-mail address: samuel.leder@icd.uni-stuttgart.de (S. Leder). Contents lists available at ScienceDirect Automation in Construction journal homepage: www.elsevier.com/locate/autcon https://doi.org/10.1016/j.autcon.2023.105082 Received 21 February 2023; Received in revised form 23 August 2023; Accepted 1 September 2023 Automation in Construction 156 (2023) 105082 2 Second, robots can inhabit the buildings they assemble giving them the ability to be involved in not only the assembly process but also in the adaptation or deconstruction process of the building. This means that the robots are required to respond to their environment and make de- cisions on the fly that can influence the design of the overall architec- tural structure as they build or inhabit it. Lastly, as the nature of CRC involves the collaboration between multiple machines, an understand- ing of these interactions and how they affect the assembly process is crucial to architectural design. The architectural design process there- fore transitions from being based on the understanding of a static workspace occupied by a single robot to a potentially unlimited work- space occupied by many robots of the same or different capabilities. Although architectural design is mentioned in previous surveys on CRC, a full review and accompanying analysis of the topic is beneficial for future research for the following reasons. First, to further evaluate the advantages of CRC systems for architecture, which include (i) working within unrestricted building envelopes, (ii) adapting to changing and particular site, design or environment conditions and (iii) assembling of complex structures enabled only through collaboration, it is necessary to understand what architectural structures can be built and in what ways they can adapt and change. Second, as systems become more complex including heterogeneous teams of robots with different capabilities, more sophisticated strategies for architectural design that factor in more parameters are required. Finally, and although the automation of the construction industry is generally slow [4], there is an increasing number of companies developing mobile robots for use on real construction sites such as BostonDynamics, DustyRobotics and Q- Bot. There are also a few companies such as hyperTunnel and OffWorld that are specifically developing CRC systems that involve multiple ma- chines collaborating to perform construction tasks. In order to make such technologies accessible to non-experts, there needs to be an un- derstanding of how architects can interact with them. Therefore, a contextualization of existing architectural design processes would be advantageous for the selection of a suitable design process in future research on CRC. This paper aims to explain the complexity of designing architectural structures for construction by MRS through a literature review. This aim is broken down into two goals: (i) to synthesize current approaches to architectural design with CRC systems and (ii) to understand the im- plications of the different approaches on the overall design to assembly process when considering CRC. The methodology and results of the re- view, compiled from literature ranging from 1997 to 2022, are pre- sented in this paper and are structured in the following manner. Section 2 clarifies terminology and discusses existing reviews on both MRS and CRC in order to explain the various other research fields and research questions involved in the development of such systems. Section 3 de- scribes the literature review methodology utilized to conduct the re- view. Section 4 introduces the categorization system developed to analyze CRC based on architectural design. Section 5 presents the various categories using exemplary research. Section 6 continues with observed trends and a discussion of the next steps for research on designing architectural structures for construction by MRS and Section 7 concludes the review with a discussion of contributions from this review. 2. Research background 2.1. Terminology Before discussing the specifics of architectural design processes in CRC, it is necessary to clarify some terminology that is used in this re- view. MRS are groups of mobile robots operating within the same environment that cooperate to achieve a complex task [3]. The robots can function autonomously, moving, acting, making observations as well as collaborating in their environment. While the term robot generally implies a physical manifestation in the real world, research based on the simulation of real robots (i.e their locomotion mechanism, or sensors are included in the simulation) is also included within this paper as to address a wider range of research. In this paper, the complex task achieved by the robots is the construction of architectural struc- tures. M