Computational approaches to space planning: A systematic review of enhancing architectural layouts

Data collection

The data collection process is formed by carefully selecting key terms, including “layout optimization,” “computational design,” and “generation algorithms.” These terms were selected to align with the research questions, ensuring that the search would cover studies on the evolution of computational tools in architectural design. Scopus and Web of Science were chosen for their extensive coverage of architecture and computational design research. This choice ensures that the review includes a wide range of studies covering theoretical progress and functional implementations, thus offering a complete investigation of the research questions. The inclusion and exclusion criteria were established to prioritize studies that were both relevant and presented a computational model, ensuring the validity of conclusions aligned with the research questions. Screening of the studies involved multiple stages, including title and abstract screening, followed by full-text evaluation. The inter-rater reliability (IRR) method was used during these stages to keep the selection process consistent and to reduce bias when choosing studies.

In order to maintain an attentive scope, only relevant articles and conference papers in English were considered valid for inclusion. The review focused on studies published between 2013 and 2023, a decade marked by significant advancements in computational design methodologies, while fields such as medicine, physics, and biology were intentionally excluded. The initial search strategy used keywords such as ‘tool’ and ‘applicat*' to test indexing accuracy. This comprehensive search string combined several related terms, yielding 7718 publications after the elimination of duplicates:

“TITLE-ABS-KEY (“space” OR “floor” OR “spatial” OR “facility”)

During the title screening phase, 694 publications were initially reviewed employing CADIMA. However, stage inadvertently excluded several previously examined publications. A second search was conducted, adding additional keywords like “method*” and “approach*” to improve the accuracy of the indexing process.

This refined search is organized to exclude papers from unrelated fields such as plantation management, space exploration, urban design, robotics, manufacturing, and electrical design, ensuring a focused analysis of computational spatial design. Keyword selection was iteratively refined based on indexing accuracy tests conducted in CADIMA and validated through independent reviewer evaluations to ensure methodological robustness. Keyword networks from both searches were analyzed using VOS viewer ⁷¹ to illustrate the differences between the initial and refined search results. Figure 4 depicts the resulting analysis, with the keywords from the first search on the left and the keywords from the second search on the right.OPEN IN VIEWER

The authors independently conducted the screening process before comparing and evaluating the results collectively. The Inter-Rater Reliability (IRR) method was employed to ensure the data’s reliability and consistency. The eligibility criteria were applied at three stages: title screening, abstract screening, and full-text evaluation. The title screening excluded studies irrelevant to the engineering fields, while the abstract screening aimed to include computational design studies relevant to architectural layout. The full-text evaluation included only articles that proposed computational generation approaches for building configurations as architectural plans and models.

This review adhered to the preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidelines, as illustrated in Figure 3. After combining search results and removing duplicates, over 13,000 publications were identified. The title screening phase reduced this number to over 10,000. The abstract screening phase further narrowed this to 296 publications, focusing on those presenting computational models for architectural space generation. Eight full texts could not be accessed, leaving 288 publications available for full-text evaluation. The inclusion criteria focused exclusively on architectural floor plans, excluding studies related to structural form-finding, landscape design, site planning, urban floor plans, retail display arrangements, and warehouse layouts. After the full-text evaluation, 82 publications met the eligibility criteria, with an additional seven records identified through citation searching and online resources. The review ultimately included 83 publications (Table 1).

OPEN IN VIEWER

Table 1.

List of 83 records by year (2013–2023).

Year	2013	2014	2015	2016	2017	2018	2019	2020	2021	2022	2023	Total
Records	72–79	80–89	90,91	92–97	98–104	105–109	110–117	118–130	131–137	138–146	147–154
Total	8	10	2	6	7	5	8	13	7	9	8	83

This process was accompanied by a high level of consistency, as demonstrated by the IRR score. The percentage agreement, or IRR, compared the number of agreements to the total number of publications reviewed. The authors reviewed 200 publications, with 182 agreements, resulting in an IRR of 91% (Table 2). This high level of agreement underscores the robustness of the literature review process, ensuring the credibility of the findings and conclusions drawn from the selected studies.

OPEN IN VIEWER

Table 2.

Inter-Rater Reliability (IRR) matrix table.

		Reviewer 2
Reviewer 1		Included	Excluded	Total
	Included	83	20	103
	Excluded	8	99	97
	Total	91	109	200

Analysis

Through a rigorous three-stage selection process based on established eligibility criteria, 83 research papers published between 2013 and 2023 on computational methods for generating architectural layouts are selected for in-depth analysis. This analysis involved creating layouts, evaluating design methodologies, testing potential solutions, and identifying gaps that required further investigation. Extracting integral data from each study involved examining several vital aspects, including article ID, title, author, building typology, input parameters, representation of spatial elements, constraints, objectives, generation models, algorithms, layout forms, dimensionality, design levels, output detail, evaluation methods, direct manipulation capabilities, user testing, research focus, and identified opportunities or arguments for employing these methods.

The literature review is systematically organized into two tables to provide a comprehensive understanding of the field. Appendix B typology identification, input parameters, regulations, and other geometrical considerations. Appendix C delves into the generation models, methods, evaluation techniques, user interaction, research focus, and opportunities identified within the selected studies. Appendix A serves as the legend for these two tables, providing abbreviations. These tables collectively offer an overview of the computational methods applied to architectural space planning over the past decade.

The study analyzed 83 selected papers to identify key variables consistently assessed across different research works. The extracted information was organized into coherent groups, and correlations were analyzed to identify emerging trends over time. This categorization facilitated a more profound comprehension of the evolution and application of computational methods in architectural design. The research included studies that employed top-down, bottom-up, or hybrid design approaches, offering a diverse perspective on the methodologies used in the field.

The systematic queries are generated to display the distribution of records based on the research questions, ensuring that the findings align with them. This approach established a direct and unambiguous connection between the extracted data and the specific research objectives. Consequently, each analysis component directly contributed to resolving the research questions. The significance of the findings was evaluated not only in terms of their contribution to existing knowledge but also in identifying what required further research. This structured analysis supports the study’s overall goal of advancing the understanding of computational methods in architectural space planning, highlighting current strengths and areas where future research can further develop these methodologies.

Methodological trends

This study has examined several procedures for generating space, including partitioning algorithms, procedural techniques like shape grammars, stochastic and evolutionary approaches, constraint fulfillment techniques, agent-based models, and, more recently, deep learning methods (Figure 5).OPEN IN VIEWER

Table 3 shows the classification of generation algorithms. Among the 83 selected studies, 24 (approximately 29%) employed procedural models such as shape grammars, while 19 (23%) studies utilized assignment models. Thirteen (16%) focused on partitioning approaches, and nine (11%) applied pixel-wise models associated with neural networks. Three studies utilized hybrid models that combined both partitioning and assignment methodologies. While 10 studies examined self-organizing models, only one had a hybrid pixel-wise self-organizing approach. ¹⁴³ Five different studies^{72,108,131,133,134} were centered around vectorial models. These studies have contributed to the development of coordinate-based movement systems.

OPEN IN VIEWER

Table 3.

Architectural space generation algorithms.

Generation model	Assignment	Partitioning	Pixel-wise	Procedural	Self-organizing	Vectorial	Hybrid
Number of records	16	12	9	23	11	5	7
Records	84,86,92,94,95,98,101,121,133,134,139-141,145,149,152	74,76,80,81,103,106,109,111,114,118,124,151	110,113,116,119,122,129,144,148,153	73,77-79,82,88-90,93,96,100,102,105,112,120,123,126,127,138,146,147,150,154	75,83,85,87,91,99,104,115,117,125,131	72,107,132,136,137	97,108,128,130,135,142,143
Representation (author)

Partitioning has declined discernibly over the last 3 years, whereas procedural, assignment, and pixel-wise models have maintained consistency. Shape grammar has been a prominent procedural model throughout the period, but this approach requires more emphasis on evaluation. Rodrigues conducts them on all four^77,78,87,88 random geometric transformation models. Assignment extensively employs fitness-based selection, where evolutionary algorithms (EA) and genetic algorithms (GA) are notably prevalent. The vectorial approach has undergone significant expansion during its evolution (Figure 6). Hybrid models mainly utilize assignment, procedural, and partitioning approaches.OPEN IN VIEWER

The research findings illustrate a development from initial procedural approaches primarily concentrating on simple rectangular configurations to more complex methods producing intricate and adaptable arrangements. The transition in methodology from rule-based to data-driven approaches addresses geometric limitations and broader functional objectives. Stochastic and evolutionary techniques introduce more randomness and variation to explore more expansive design spaces creatively. However, they need to possess accuracy to meet the specified limitations. Most studies aim for full automation without direct manipulation during the design generation process, which causes insufficient utilization of architect creativity and expertise. While direct interaction remains limited during production, agent-based systems offer indirect control by incorporating user goals through agent behavior specifications. Designers can customize automatically generated plans to incorporate human creativity through filters. Deep learning techniques focus more on automated precision rather than directly supporting creative control. However, they can rapidly generate numerous design alternatives, enabling qualified and diverse explorations. Recent studies have utilized optimization and deep learning to enhance performance and generalization abilities beyond fundamental constraints (Figure 7). Hybrid methods mostly combine graph-based, shape grammar, simulation-based, Voronoi diagram, and AI algorithms.OPEN IN VIEWER

Handling multi-objective design criteria and trade-offs

In architectural design, balancing multiple objectives is a complex challenge that often involves trade-offs between competing criteria. These multi-objective design problems require detailed consideration of various geometric, topological, and functional constraints. The goal is to find solutions that fulfill these often-conflicting criteria while still adhering to the overall design vision. This section explores the strategies and methodologies used to handle multi-objective design criteria.

A comprehensive understanding of the various constraints governing design processes is fundamental to effectively addressing RQ2. Almost all studies focus on fulfilling geometric and topological rules while preserving unit sizes, ensuring adjacency, and adhering to non-overlapping requirements. Comprehending the spatial relationships is the initial goal, regardless of whether the study pertains to residential, educational, or office spaces. Geometry, topology, and goal, as outlined in, constitute the three primary types of constraints and objectives that studies encounter.

Table 4, constitute the three primary types of constraints and objectives that studies encounter.

OPEN IN VIEWER

Table 4.

Classification of constraints and objectives.

Records	Combination	Geometry								Topology			Goal
		Number	Area	Size-percent	Direction & position	Distance	Shape	Non-overlapping/ overflowing	Opening	Adjacency & connectivity	Based on dataset	Shape rules	User-specified	Type & function	Lighting & shadow & solar	Circulation	Cost	Other
74,81,82,88,90,91,94,100,101,107,109,113,114,116,120-122,124,127,128,133,139,141,147,148,151	GeometryTopologyGoal	X	X	X	X	X	X	X	X	X	X	X	X	X	X	X	X	X
75,76,78,79,83,84,86,89,92,93,95,97-99,111,112,119,125,130,135,136,138,140,149,153,154	GeometryTopology	X	X	X		X	X	X	X	X	X	X
85,110,117,144,145,150,152	TopologyGoal									X	X	X		X	X	X	X
72,73,77,80,103,108,115,132,137,146	GeometryGoal	X	X	X	X	X		X	X					X	X	X		X
87,104,105	Geometry	X		X	X	X		X	X
118,131,142,143	Goal												X		X			X
96,102,106,123,126,129,134	Topology									X	X	X

Inputs and unit representation are examined through various methodologies, including graphs, matrices, and visual models, to explore the representation and management of constraints. Spatial geometries are constructed using rectangular duals, voxel grids, continuous cells, and pixel-wise representations. The outputs consist of various visual representations, beginning with simplified block shapes and advancing to more complex designs that include room labels, furniture, textures, and other pertinent information. Furthermore, common elements in unit representation such as ‘point coordinates,” rectangle,’ and 'colors and textures’ suggest a universal understanding and straightforward interpretation across various contexts (Appendix B). Multiple studies utilizing graph representations have demonstrated the effectiveness of these representations in maintaining spatial relationships and constraints. Most studies focus on residential layouts, typically presented as rectangular, two-dimensional, and simplified diagrams. These geometrical and visual diagrams represent the distribution of layout shape, size, level, detail, and representation (Figure 8).OPEN IN VIEWER

Evaluating the effectiveness of algorithms and methodologies in managing multi-objective criteria requires focusing on their ability to balance competing design objectives while adhering to established constraints. For example, designers widely employ evolutionary algorithms to explore the trade-offs between competing objectives. These algorithms allow for iterative optimization, where designers can weigh different criteria according to their priorities. Additionally, deep learning methods, such as generative adversarial networks (GANs), are increasingly used to generate detailed designs that meet specific objectives at both 2D and 3D levels. They enable the rapid generation of design alternatives, but their limited ability to incorporate user feedback highlights the need for hybrid approaches that balance automation with creative control. Shape grammar and agent-based models let the designer procedurally and interactively use rule-based and behavior-driven criteria. These models can handle the trade-offs between circulation patterns, spatial efficiency, and designer preferences.

Performance-oriented design goals introduce another layer of complexity. Studies focusing on cost, lighting, and thermal constraints often employ bottom-up approaches, allowing for emergent solutions that balance various design criteria. Table 5 summarizes the various performance-oriented design approaches utilized in 46 selected studies, providing a clearer understanding of how these methods contribute to managing multiple criteria seamlessly. Bottom-up approaches allow for emergent solutions driven by localized rules, providing flexibility in achieving numerous design criteria. In contrast, hybrids of bottom-up and top-down tactics balance detail-oriented and holistic design goals. Top-down techniques strive to accomplish overarching goals through hierarchical design processes, guaranteeing an organized approach to all objectives.

OPEN IN VIEWER

Table 5.

Performance-oriented records.

	Approach
Generation model	Bottom-up		Combination		Top-down		Total Number of Records
Assignment	3	101,133,152	3	139,141,145	2	94,121	8
Assignment; partitioning			2	108,128			2
Assignment; procedural			1	142			1
Partitioning					9	74,80,81,103,109,114,118,124,151	9
Pixel-wise	3	110,144,148	3	113,116,122			6
Procedural	6	82,88,90,120,127,147	4	77,100,146,150			10
Self-organizing	5	85,91,115,117,131					5
Self-organizing; pixel-wise	1	143					1
Vectorial	3	107,132,137			1	72	4
Total Number of Records	21		13		13		46

Managing multi-objective design needs and trade-offs is essential for achieving balanced, functional architectural outcomes. Designers can manage the intricacies of multi-objective challenges by combining advanced visual representation techniques, computational approaches, and performance-oriented strategies to arrive at solutions that meet both restrictions and design goals.

Evaluation, benchmarking, and practical application

Evaluating automated architectural space planning models plays a significant role in understanding their functionality and applicability within practical contexts. This procedure commonly entails evaluating and comparing reference layouts, evaluating their compliance with stated restrictions, and examining optimization metrics. Furthermore, the qualitative feedback from consumers provides crucial insights into the usability and feasibility of the developed designs. Assessment methods commonly involve a combination of quantitative and qualitative approaches. Quantitative measures prioritize factors such as meeting constraints, achieving optimization goals, maximizing spatial efficiency, and evaluating daylighting and thermal performance. Comparative benchmarking compares several approaches, providing a distinct understanding of each model’s performance in certain circumstances (Table 6).

OPEN IN VIEWER

Table 6.

Evaluation methods and metrics used.

Evaluation aspect	Evaluation method	Metrics
Quantitative metrics	Constraint satisfaction	Degree of adherence to specified constraints
	Optimization objectives	Achievement of optimization goals (e.g., minimizing cost, maximizing efficiency)
	Similarity to reference layouts	Quantitative comparison with predefined reference layouts
	Spatial efficiency	Space utilization, functionality, distribution, and compactness
	Daylighting/Thermal performance	Quantitative assessment of natural lighting and thermal comfort in the designed spaces
	Travel distance/ Circulation efficiency	The distances required to move and the efficiency of movement paths
	Comparative benchmarking	Benchmarking between different methods
Qualitative methods	Interviews	In-depth conversations with users to gather insights on their experiences and needs
	Questionnaires	Structured surveys to collect feedback and opinions on various design aspects
	Expert reviews	Assessment by domain experts focusing on usability, user experience, and creativity support
	Visual assessment	Evaluation through visual inspection of generated configurations
	User feedback	Collection of feedback from designers on the aesthetic appeal and quality of the configurations
Design analysis	Simulation tools	Structural integrity, energy consumption, accessibility, acoustics, visual aesthetics, etc. of the generated configurations using simulation tools
	Comparison	Plausibility through comparison with real architect-designed layouts
	User studies	Usability, quality, and plausibility through ratings from architects

The persistent use of quantitative methodologies, including GA and EA, highlights their significance in assessing architectural design models throughout time (Table 7). Nevertheless, the data does not show a substantial increase in the utilization of quantitative methods, implying a consistent preference for these approaches. Earlier studies primarily observed qualitative research approaches like user feedback and expert reviews, suggesting a potential area for improvement in future studies.

OPEN IN VIEWER

Table 7.

Distribution of evaluation methods over years.

Evaluation	Type	2013	2014	2015	2016	2017	2018	2019	2020	2021	2022	2023	Total Nuber of Records
Evolutionary Algorithms (EA)	Qt	1						1					2
Firefly Optimization (FO)	Qt						1						1
Genetic Algorithms (GA)	Qt	1	1		1	2		1		1	1	2	10
Linear Programming and Optimization (LP&O)	Qt								3				3
Benchmarking, metrics	Qt							3	5		4	3	15
Multi objectiveGenetic Algorithms (MOGA)	Qt					3		1	1	2	1		8
Search Optimization (SO)	Qt					1							1
Simulated Annealing (SA)	Qt	1			2								3
Simulation-Based (SB)	Qt		1							1			2
Space Syntax (SS)	Qt	1		1									2
Swarm Intelligence (SI)	Qt		2										2
User Feedback and Selection (UF&S)	Ql	2	2	1	1		1	1				1	9
Hybrid	Qt/ Ql	2	2		1		2	1		2	3		13
Subtotal Number of Records (qualitative)		2	2	1	2		1	2		1	1	1	13
Subtotal Number of Records (quantitative)		6	6	1	3	6	3	6	9	5	8	5	58
Subtotal Number of Records (No evaluation)			2		1	1	1		4	1		2	12

A primary challenge in evaluating computational models is the need for standardized datasets and tasks, which hinders the ability to benchmark and compare findings from various studies accurately. Furthermore, directly comparing generative approaches and algorithms is difficult because of procedural disparities, and the presence of human contribution can further confound comparisons. Measuring algorithmic competence necessitates using specific metrics, such as computation time which reflects how quickly the algorithm generates a complete and optimized layout, and accuracy of results, which evaluates how sufficiently the generated layouts meet predefined constraints, such as adjacency or circulation flow quality. However, finding a single optimal solution is challenging due to the conflicting optimization goals. A comprehensive evaluation methodology with diverse indicators and approaches would enable a more thorough evaluation of these models.

Moreover, creating representative datasets verified using traditional measurements and procedural standards could facilitate the implementation of standardized evaluation methods. Weber et al., ⁶⁸ propose a list of measurements for automated spatial organizations that encompass spatial, environmental, and structural aspects. These metrics can improve space use and functionality in modularity, compactness, and adaptability. Standardized datasets, such as those proposed by Weber et al., could facilitate consistent benchmarking, allowing for more reliable study comparisons.

Research gaps and future directions

Nine focus areas categorize current computational methods for architectural space planning: form and design, efficiency and performance, configuration, methodology, design tool, constraints, reuse, user, and benchmark (Table 8). Procedural models primarily emphasize ‘form and design,’ whereas vectorial and partitioning models prioritize ‘configuration.’ On the other hand, assignment and self-organization models concentrate on ‘constraints’ and ‘efficiency and performance.’ Lastly, pixel-wise AI models center their attention on ‘methodology.’ These computational methods have revolutionized spatial organizations by offering diverse approaches to complex design challenges. Each category addresses specific aspects of the planning process, from aesthetic considerations to functional optimization. By combining the strengths of different approaches, planners and designers can create more adaptable, intuitive, and environmentally conscious spaces. As technology advances, further refinement of these methods is expected, potentially leading to more holistic architectural space planning solutions. The main research can be addressing as user interaction and experience, modularity and contextual knowledge and educational use and underexplored aspects.

OPEN IN VIEWER

Table 8.

Focus, Generation model and user experience matrix.

Focus	Generation model							User exp
	Assignment	Partitioning	Procedural	Self-organizing	Vectorial	Pixel-wise	Combination	Interaction and test
Form and design	139,152	74	73,120,146	99,104	136			73,74,146,152
Efficiency & performance	98,141,145	118	88,147	87			97,108	97,98
Configuration	86,95,133	76,81,114	89	75,115,125	107	110,148		75,107,148
Methodology	94,140	106,109,111,151	78,90,93,100,102,105,154			113,116,119,122,144	143	90,93,109,143
Constraints	92,149	124	79,127		137		128,130,135,142	92
Design tool			112,123	83		129		83,112
User	84,121,134	80	96	85,91,131		153		91,96,121,134
Reuse			150	117
Form and design combination			126
Efficiency & performance combination	101	103	77,82,138		72,132			72,132

The review reveals a significant gap in interaction within computational tools, with 84% (70 out of 83) of studies. This finding underscores a heavy reliance on theoretical or computational evaluations rather than empirical validation. Most computational tools lack mechanisms for incorporating user feedback, reducing their relevance and potential for participatory design.

Among the studies that included interaction, expert tests were the most common, appearing in six records across various years. Expert feedback provides valuable insights, particularly in complex or specialized domains, yet remains underutilized. There were only four instances of pilot tests involving users or students. Two studies that combined student and expert tests observed a more comprehensive validation approach, though such examples could have been more frequent (Table 9). These participatory methods, exemplified by Azadi et al.'s ¹³¹ and Fernando’s ⁸⁴ approaches, highlight the untapped potential for accessible design. Incorporating interaction more systematically into computational tools could enhance their relevance, fostering more outstanding adoption in field implementation of these techniques.

OPEN IN VIEWER

Table 9.

Distribution of the user tests in reviewed papers.

Interaction & user test	2013	2014	2015	2016	2017	2018	2019	2020	2021	2022	2023	Total Number of Records
Expert test			90	92,97	98						148,152	6
Pilot user test	74			93					134			3
Student test				96								1
Student test; expert test									132	146		2
Survey							112					1
Without user test	7	10	1	2	6	5	7	13	5	8	6	70
Total Number of Records	8	10	2	6	7	5	8	13	7	9	8	83

Table 10 indicates that current methods experience several significant issues. Firstly, these methods prioritize algorithmic performance and disregard examining the balance between configuration quality and computational performance. Additionally, these methods lack seamless interfaces that bridge the gap between manual and automated systems. Among the few studies that address these challenges, notable examples include Upadhyay et al., ¹⁵² aiming to simplify the design process without requiring specialized expertise or tools, and Bahrehmand et al., ⁸⁰ focusing on streamlining the architectural workflow by enabling iterative input adjustments. These approaches highlight the potential for more user-centric and adaptable design methodologies. Similarly, Azadi et al. ¹³¹ emphasize participatory design through customizable shape selections and user polls, while Garip et al. ¹³⁴ provide customizable options via straightforward user interactions. Heckmann et al. ¹²¹ offer a system for customizing residential building layouts based on preferences, and Fernando ⁸⁴ uniquely provides interaction through sketching, offering a more intuitive design interface.

OPEN IN VIEWER

Table 10.

Challenges and limitations for automated spatial organization approaches.

Challenge	Keywords	Limitations
Methodological	User/human	The predominant focus is on algorithmic performance
	Configuration quality	Insufficient exploration of tradeoffs with computational efficiency
	Modular	Preference for custom-coded systems
	Interaction	Lack of a seamless collaborative interface between fully manual and fully automated systems
	Standardization	Evaluation challenges
Design-related	Automation efficiency	Constrained by primary architectural design rules and objectives
	Contextual knowledge	Insufficient integration of environmental data
	Design workflow	Disjointed phases of the workflow
	Real-world project	Challenges in verifying or confirming
	Accessibility/health	Lack of research on human health outcomes, accessibility, universal design, and inclusiveness
	Re-use	Limited exploration of constrained spaces in preexisting structures

While considerable studies prioritize computational efficiency and algorithmic performance, they often underemphasize human-centered approaches. This focus on algorithmic metrics overlooks the implications for end-users and makes it more challenging to validate computational methods, as highlighted in Research Question 3. The need for standardized evaluation processes further complicates this issue, leading to inconsistencies in assessing the usability of generational models and tools.

Many studies focus on design exploration and the development of computational tools, yet they often struggle with balancing computational efficiency and the quality of configurations. Although optimization and constraint handling opportunities are explored, the balance between these computational goals and the usability or effectiveness of designs from a human perspective has yet to be investigated. This limited focus on participation, especially in human-centered and participatory design areas, underscores the need for more comprehensive user studies to ensure that computational tools are applicable and intuitive. Another highlight of user experience is the impact of the COVID-19 pandemic, which is evident in the research trends observed in 2020. There was increased emphasis on theoretical and computational methods and decreased empirical testing. This shift likely reflects the challenges of conducting in-person studies during the pandemic. However, it also underscores the continuous requirement for more consistent empirical validation in future research.

The preference for custom-coded systems over modular, reusable toolkits exacerbates the challenges in computational design methodologies. This trend limits the reproducibility and applicability of results in operational scenarios. Custom-coded systems lack adaptability and scalability, creating barriers to combining manual and automated processes. The analysis highlights a pronounced inclination toward custom-coded systems in computational spatial design, even though modularity and standardization in tool development are fundamental to achieving widespread adoption. For instance, designers can create and share modular components like Planmaker ¹⁵⁵ and Magnetizing FPG ¹⁵⁶ for various design tasks using Grasshopper, ¹⁵⁷ a visual programming tool for Rhino.

Similarly, Dynamo, ¹⁵⁸ embedded into Autodesk Revit, provides a node-based interface where designers can build and customize workflows for parametric designs. With these toolkits, designers can combine pre-built components or develop custom scripts, exemplifying modularity and maintaining consistency across projects. They offer considerable potential for improving collaboration and innovation, particularly with contextual knowledge and workflow optimization advancements.

Another significant shortcoming of the analyzed approaches is their insufficient incorporation of the environmental context, such as topography and neighboring areas. The studies often overlooked the context, limiting their adaptability to applications. While direct evidence of enhanced usability is limited, the reviewed studies highlight how tools with environmental data facilitate more context-aware designs, potentially leading to improved user experiences and functional adaptability. Although this study identifies the potential of environmental data for enhancing usability and adaptability, empirical validation remains limited.

Integrating computational design in architectural education not only enhances students’ technical skills but also encourages critical thinking and problem-solving abilities. As noted by Oxman, ¹⁵⁹ these digital tools can serve as a medium for design exploration, enabling students to engage with complex geometries and performance-based design strategies. Furthermore, using parametric modeling and generative design techniques, mainly through scripting, equips students with the evolving demands of architectural practice, where such tools are increasingly prevalent. ¹⁶⁰

Despite these advantages, the literature review analysis reveals another notable research gap: there is a need for more emphasis on the educational applications of computational spatial design tools. Three studies applied user tests on students.^96,132,146 Further research is needed to explore how these increasingly sophisticated and advanced tools can be effectively introduced and utilized within architectural education. This gap is particularly significant given the growing implications of digital literacy and computational thinking in architectural contexts. Architecture students should focus on using these tools to learn design principles and improve their spatial organization skills. Incorporating them into educational programs could significantly improve student outcomes, preparing future architects to navigate the profession’s technological and methodological advancements.

While educational applications of computational tools are critical, their more expansive implications in addressing underexplored areas like accessibility and inclusiveness also demand attention. These shortcomings suggest a need to explore how computational tools can enhance inclusiveness in design, particularly for underserved user groups. Research in this area could also equip architecture students with the skills to develop inclusive design strategies, fostering collaboration and innovation in addressing the needs of underserved user groups.

Future research should prioritize bridging these gaps by developing intuitive and user-friendly strategies. These approaches should include diverse and frequent user testing, particularly involving end-users and students, to enhance the robustness and applicable relevance of the research outcomes. Employing a combination of theoretical, computational, and empirical validation methods will improve algorithmic performance while aligning with the practical needs of architectural projects. Such efforts will enhance these tools’ theoretical soundness and practical utility, addressing the varied demands of the architectural community and educational institutions.

This research builds upon earlier works, such as those by Homayouni, ⁶² Lobos and Donath, ⁶⁴ and complements specialized studies by Weber et al. ⁶⁸ and Veloso et al. ⁶³ While documented surveys by Homayouni, Lobos and Donath provided a detailed account of the evolution of computational methods and challenges from the 1960s to the early 2000s, this study focuses on the advancements that occurred between 2013 and 2023.

One of the distinctive findings is the increased capability of computational tools to handle multi-objective design criteria, which involve balancing various aspects, such as spatial relationships, environmental context, and user preferences. This observation aligns with the findings of Weber et al. and Veloso et al., exploring specific methodologies that enable the simultaneous management of multiple objectives. Both emphasized the importance of advanced algorithms that enhance the precision and applicability of design tools. The findings support these insights by directing their capabilities into intuitive and scalable systems.

This study reaffirms the persistent challenges in embedding computational tools within architectural design workflows, as emphasized by previous research. Their limited ability to accurately address the complexities of professional practice remains a significant barrier to incorporation. Lobos and Donath’s critique remains pertinent, but recent developments suggest improvements. Adopting computational tools is more feasible when designed to complement and enhance creative processes rather than constrain them. Veloso et al. noted that current user-centered tools offer more flexibility and adaptability and align with architects’ needs, emphasizing the importance of accessible and intuitive tools.

While Lobos and Donath were skeptical about the widespread adoption of computational tools due to their limited alignment with the utilization demands of architectural practice, the findings of this study suggest a more optimistic outlook. Recent improvements in user engagement and environmental parameters enhance the potential for these tools to become an integral part of routine architectural workflows. This optimistic perspective aligns with Weber et al.'s findings, highlighting how advanced computational methods can complement design's creative aspects rather than replace them. Such tools support creativity and integrate contextual and performance-driven requirements into the design process.