MQue Hosts Roundtable: How 3D Generative AI is Expanding Architectural Design with Obayashi Corporation and Mitsubishi Jisho Design

(From left) Kai Katsumata, Towaki Takikawa, Aki Sato

MQue, a deep tech company specializing in the development and social implementation of unique technologies in simulation, modeling, and conversational AI (CEO: Takuya Tsuda), held a roundtable event on "Exploring the Possibilities of 3D Generative AI in Architectural Design" on June 29, 2026 (Mon). The event, titled "Architecture x AI Roundtable (2nd Session) '3D Generation: Expanding Ideas,'" took place at the headquarters of Mitsubishi Jisho Design Inc.

MQue is a deep tech partner that combines multiple cutting-edge technologies developed by researchers with world-class achievements to implement them in practical applications. In its modeling business, MQue's core research area is computer vision, with a foundation in technologies that accurately understand and structure visual information such as images, drawings, and 3D data.

This roundtable was planned as an invitation-only, closed-door event to share and organize the practical applications of 3D technologies, including 3D generative AI, in the field of architectural design, against the backdrop of their rapid advancement. The second session focused on "3D Generation: Expanding Ideas," with the goal of elevating the perception from "3D generation is an interesting technology" to "3D is an area where we should identify potential applications in our own operations," enabling each company to clearly take home potential uses for their respective businesses. On the day, practitioners from Obayashi Corporation and Mitsubishi Jisho Design Inc., involved in design, DX, structural engineering, construction, and management strategy, gathered to discuss the following points:

When 3D spaces and videos can be generated from prompts, sketches, and images, who within a company can best utilize them?

How do applications beneficial for architectural design differ from those beneficial for facilities and construction?

What kind of meaning should be assigned to 3D data to enable its utilization in actual operations?

This report summarizes the lectures and key discussion points from the event, sharing the current state of architectural design amidst the advancement of 3D generative AI and insights for future practical applications.

Lecture by Towaki Takikawa: "What is 3D? Systematizing 3D Generation Technology by Input and Output"

In the first half of his lecture, Towaki Takikawa started by asking, "What is 3D in the first place?" and "Why is 3D data necessary?" He explained that the required technologies for 3D vary greatly depending on the objective, and presented a systematic way to understand the capabilities of 3D technology by categorizing it based on combinations of "input" (text, sketches, images, generated 3D spaces, etc.) and "output" (videos, 3D spaces, 3D for analysis/verification, BIM, etc.).

He incorporated a wide range of the latest technological trends, including research examples of generating 3D spaces and experiential videos from text, sketches, and images. He concluded that by adopting a modular perspective that considers combinations of input and output, it becomes possible to identify which of a company's operations can benefit from these technologies.

Lecture by Kai Katsumata: "From Scanning Real Spaces to 3D Reconstruction and Connection to Operations"

In the latter half of the lecture, Kai Katsumata explained the process of scanning real spaces, reconstructing them in 3D, and connecting them to business operations. He introduced the latest research trends, including the distinction between "measurable 3D" that accurately preserves dimensions and "visual 3D" that appears realistic like a photograph, acquired through LiDAR or photography. He also discussed the concept of adding meaning and attributes to 3D data to connect it to BIM and drawing operations.

He emphasized that "visual appeal" and "measurement accuracy" are distinct, and it is crucial to determine the necessary accuracy based on the purpose of the judgment. Numerous concrete points for implementation in the field were shared.

Q&A and Panel Discussion

Following the lectures, participants engaged in a Q&A session, with Mr. Takikawa and Mr. Katsumata responding in a panel discussion format. Key discussion points included:

The current state and accuracy of technologies for scanning real spaces and converting them into 3D.

The concept of assigning meaning to 3D data and connecting it to operational data such as drawings and BIM.

Points to consider when utilizing 3D across different processes, from design to construction.

How to bridge the gap between advancements in cutting-edge technology and practical field operations.

The discussion highlighted the importance of starting with the purpose – "for what judgment, and with what accuracy should it be used?" – rather than solely focusing on what 3D technology can do. Furthermore, positive prospects were shared regarding the potential to leverage the accumulated data unique to the architectural field to adapt generally evolving technologies to specific company operations.

Group Discussion: "Exploring the Potential of 3D in Our Company's Operations"

In the latter half of the event, group discussions were held at each table on the theme of "Where can 3D be utilized in our company's operations?" It was shared that the requirements for 3D differ depending on the role and profession involved in design, and that the value of the same technology changes depending on "who uses it, in what situation, and for what purpose."

The session fostered passionate dialogue, with participants concretely considering the application of 3D to their own company's operations, thereby enhancing their understanding of the next steps toward implementation, such as "If the technology has advanced this far, we would like to use it in this way."

Conclusion: Towards the Implementation of Architecture x 3D Generative AI

3D technologies, including 3D generative AI, are steadily moving towards the implementation phase in architectural design. However, as the range of technological options expands, the practical judgment of "which technology to use, for what purpose, and to what extent" becomes increasingly important.

This roundtable systematically organized the capabilities of 3D technology along the axes of input and output, and by discussing them in relation to each company's specific operations, a common perspective was shared for addressing the question, "In which of our company's operations does 3D have potential benefits?"

It is becoming increasingly important to proactively view 3D generative AI not just as an "interesting technology" but as an "area where we should identify its benefits for our own operations."

MQue will continue to aim for the reliable integration of cutting-edge research results, including computer vision, into practical architectural design. This includes not only technological development but also creating forums for discussion that encompass the criteria for utilization.

Comment from Towaki Takikawa

By organizing 3D technology not just by "what it can do" but by "what it can be used for" through combinations of input and output, it becomes clear which of a company's operations can benefit. Much of the research is conducted openly, and I believe that by pooling knowledge, implementation in the architectural field will steadily advance. I hope today's discussion will serve as a seed for that.

Comment from Kai Katsumata

Technologies for scanning real spaces, reconstructing them in 3D, and connecting them to drawings and BIM have advanced significantly in the past one to two years. However, "visual appeal" and "measurement accuracy" are distinct, and it is important to determine the necessary accuracy based on the purpose of the judgment. Discussing this in relation to everyone's specific tasks was a valuable opportunity to bring research closer to practical application.

Comment from Aki Sato, MQue Inc.

I felt a great sense of accomplishment from seeing developers at the forefront of research and practitioners at the forefront of architectural design engage in frank discussions at the same table. We witnessed many moments where the "use cases" that are not apparent from technology discussions alone or field discussions alone became concrete when the two perspectives converged. Starting from this dialogue, we want to connect research results to field implementation one by one.

Event Overview

Event Name: Architecture x AI Roundtable 2nd Session "3D Generation: Expanding Ideas"

Date & Time: June 29, 2026 (Mon) 9:00 AM - 12:00 PM

Venue: Mitsubishi Jisho Design Inc. Headquarters

Participating Companies: Obayashi Corporation, Mitsubishi Jisho Design Inc.

Speaker Profiles

Towaki Takikawa

Founder and CEO of Outerport. Originally from Oregon, USA, he focused on robot design and metal processing during high school. After completing his computer science degree at the University of Waterloo, he is currently on leave from his PhD program at the University of Toronto. He joined NVIDIA as a researcher, engaging in R&D for computer vision and simulation. His numerous papers have been accepted at leading international conferences such as CVPR and SIGGRAPH, with cumulative citations exceeding 5,000. He later founded Outerport, which was accepted into Y Combinator. He is tackling the severe labor shortage in the manufacturing and engineering industries with his "Design AI Agent," which reads drawings and technical documents for electrical and plumbing systems, and handles everything from the design of new equipment and plants to parameter optimization.

Kai Katsumata

Currently pursuing a PhD at the Graduate School of Information Science and Technology, The University of Tokyo, where he researches 3D Gaussian Splatting, GANs, image generation/editing, and domain generalization. His research on dynamic 3D scene representation was accepted at ECCV 2024, and his research on GAN inversion and conditional image generation was accepted at WACV 2024. He has also presented research findings at international conferences such as CVPR, ICIP, and ICASSP. He is currently engaged in cutting-edge research in the 3D/image generation AI domain, including dynamic Gaussian Splatting and research on image generation and visual artifact removal.

Aki Sato

Engaged in production technology at Sumitomo Electric Industries, where she worked on improving the productivity of existing products, setting up production lines for new products, and quality improvement. Subsequently, at McKinsey & Company, she supported manufacturing and consumer goods companies in strategy formulation, operational transformation, and sales reform. At MQue, she is responsible for organizing customer challenges, implementation design, and business promotion in the modeling business, which handles drawing checks and 3D technologies.

About MQue Inc.

MQue Inc. is a deep tech company that develops and socially implements unique technologies in simulation, modeling, and conversational AI.

We address societal challenges by connecting cutting-edge technologies from researchers with world-class achievements to high-difficulty problems faced by companies and organizations.

We are confronted with the issue that advanced technologies and expertise are often tied to individuals or organizations, becoming siloed and not fully utilized by society as a whole. Particularly in fields requiring complex optimization and exploration, decision-making occurs with fragmented principles related to design, performance, and human insight. As a result, in manufacturing, architecture, and organizational management, decisions that are only locally optimal or reliance on individual expertise make it difficult to achieve overall optimization and ensure reproducibility.

MQue's integrated vision is to derive optimal solutions that balance feasibility, performance, and acceptance by unifying design, performance, and human insight into a single decision-making process.

We emphasize not just providing technology but integrating research results into actual operations and decision-making, carrying through to social implementation.

To realize this vision, we develop research and business integrally in three areas: "Simulation," "Modeling," and "Conversational AI."

Business Overviews

Simulation Business

We are working to advance manufacturing, led by researchers with world-class achievements in fluid simulation, AI surrogate models, and optimization. In collaboration with companies in the thermal equipment and aerospace industries, we provide not only simulation development and application but also the construction of surrogate models (1D-CAE models, AI surrogate models) and optimization based on them. By effectively combining experiments and simulations, we achieve faster design processes and improved performance while reducing the costs of prototyping, etc. This enables the use of simulations in design judgments and decision-making, allowing for optimization across design and performance.

Modeling Business

We are optimizing design processes in the architecture and manufacturing industries in collaboration with engineers who have received high acclaim at CVPR, the world's premier computer vision international conference. For general contractors, developers, and architectural firms, we provide automation and optimization of design drawing checks. For manufacturers in the architecture and manufacturing sectors, we support enhanced proposal and design capabilities that reflect customer preferences and design intentions. By structuring processes that have relied on the experience and judgment of designers, we support decision-making that considers both human insight and design.

Conversational AI Business

Researchers in human-AI collaborative intelligence and consultants specializing in organizational design and executive development collaborate to develop our unique conversational AI technology and use cases. In collaboration with major financial institutions, general trading companies, and HR-related companies, we are rebuilding personnel allocation, evaluation, training, and recruitment, which have traditionally relied on intuition and experience, based on scientific methods and company-specific contexts. By utilizing unstructured data from "conversations" to visualize human tacit knowledge and decision-making processes, we support the advancement of decision-making, including human insight.

Company Overview

Company Name: MQue Inc.

Address: Sumitomo Fudosan Iidabashi Building 4F, Room 8, 2-3-21 Koraku, Bunkyo-ku, Tokyo

Representative: Takuya Tsuda

URL: https://mque.co.jp/

<Inquiries Regarding This Matter>

MQue Roundtable PR Office: Takahashi

Email: [email protected]