CDFAM Computational Design Symposium

CDFAM Computational Design Symposium — Tokyo 2026

Kotaro Tanimichi · Nature Architects

Nature Architects Inc. develops a software platform that deeply integrates geometry processing, numerical analysis, and optimization, while also providing design solutions enabled by the platform through its Generative Engineering™ approach. Covering a wide range of manufacturing contexts, from mass production to advanced manufacturing methods, the company dramatically accelerates design exploration and creates new designs beyond existing design paradigms.

This presentation will introduce case studies in which design concepts were generated under complex, large-scale physical phenomena and geometric constraints. Examples include the rapid exploration and creation of new design concepts for electric vehicle bodies, innovative garment design using origami-based techniques, and the design of highly efficient heat exchange structures enabled by additive manufacturing. The presentation　will also outline future directions for Generative Engineering™

Links

Talk page with full transcript

Watch the talk on YouTube

Search this talk's transcript

Cite this talk

CDFAM Computational Design Symposium — Tokyo 2026

Zinab Eisa & Esha Morakhiya · Harvard University

Aeroforms emerges from a shared curiosity of how the invisible sensory phenomena of scent can be modelled, mapped and ultimately understood as data, with a unique focus on scentscapes of Japan.

Scent is one of the most powerful triggers of memory and belonging, yet it remains largely absent from environmental modeling and urban analytics. In 2001, Japan’s Ministry of Environment created “100 Scents of Japan”, a curated archive of culturally significant smellscapes tied to regional identity. The catalog revealed how smell functions as a vessel of ecological collective memory. Two decades later however, the physical landscapes that produced these scents have changed. Industrialization, land-use transformation, and air pollution have altered Japan’s atmospheric composition, and with it, its olfactory ecology.

Aeroforms surfaces at the intersection of environmental science and design engineering to address this gap by asking: how can we computationally model scent at a national scale, and what happens when we compare modeled atmospheric composition to cultural memory?

We developed a computational framework for olfactory modeling, a way to approximate the composition of scents across Japan using quantitative environmental data. Through this framework, we constructed a sensory digital twin of Japan’s olfactory environment, integrating agricultural and industrial datasets to estimate the relative presence of ten scent taxonomies across all prefectures. We derived emission factors from literature-based odor threshold values using inverse-potency summation, median log compression, and adaptive normalization. These were then multiplied by prefectural land area to estimate natural emissions, with industrial pollutant emissions incorporated as a parallel weighted contribution layer. The result is a combined, normalized composition vector per prefecture.

By comparing these modeled scent compositions with the “100 Scents of Japan” catalog, we measure where cultural memory aligns with — and diverges from — ecological and industrial reality. This research therefore operates at the intersection of environmental data science, cultural geography, and sensory humanities, developing a reproducible model for environmental perception and introducing a methodology for quantifying perceptual-environmental divergence or, the measurable difference between how people remember environments and how those environments now exist materially.

Links

Talk page with full transcript

Watch the talk on YouTube

Search this talk's transcript

Cite this talk

CDFAM Computational Design Symposium — Tokyo 2026

Laurence Cook + Beatrice Zanardi · Generative Engineering

What could the next generation of engineering software feel like, and what becomes possible when the interface is no longer the constraint?

The tools we use to create are changing. AI is making it possible to move beyond traditional software interfaces, but the opportunity isn’t simply to add AI to the interfaces we already have. It’s to rethink the interaction layer itself. Explore what happens when the interface becomes more fluid, more exploratory, and more directly connected to the human act of engineering

Links

Talk page with full transcript

Watch the talk on YouTube

Search this talk's transcript

Cite this talk

CDFAM Computational Design Symposium — Tokyo 2026

George Allen · nTop

By now, implicit modeling technology and its benefits are well understood. In situations where a part is highly complex or its design process is highly iterative, implicit modeling using signed distance fields is uniquely able to deliver the reliability, performance, and scalability that are required for success. However, the novelty of implicit modeling is also a source of problems. Current engineering/manufacturing processes are often heavily dependent on legacy software tools that cannot consume implicit models — they require meshes or boundary representations, instead. So how do we address this problem? There are a few options — we can either convert implicit models to the forms needed by legacy tools, replace the legacy tools with new ones that consume implicit models directly, or change the data flow in the overall process. The talk examines these issues, reports progress and the current status, and suggests some ways forward.

Links

Talk page with full transcript

Watch the talk on YouTube

Search this talk's transcript

Cite this talk

CDFAM Computational Design Symposium — Tokyo 2026

Darren Henry · PTC

The race toward AI-generated CAD has largely focused on teaching models to produce geometry directly—from B-reps and meshes to feature histories and sketches. But what if we’re teaching AI the wrong language?

This presentation explores an alternative paradigm: Text-to-Code-to-CAD, where large language models generate engineering logic rather than geometry itself. Instead of predicting low-level geometric entities, AI writes executable CAD code that captures design intent, constraints, parameters, and engineering relationships. The result is not merely a finished model, but a native, editable, and reusable engineering feature that integrates naturally into the design workflow.

Using real-world examples, including AI-generated CAD features built through FeatureScript, this session demonstrates how modern language models can write, test, debug, and refine engineering code to create complex parametric designs while preserving the transparency and editability required for professional engineering. The discussion contrasts this approach with direct geometry generation and examines why engineering intent—not geometry—may be the more scalable representation for AI-assisted design.

Attendees will leave with a new perspective on computational design: the future may not be AI replacing CAD, but AI leveraging programming languages purpose-built for engineering to accelerate design while keeping engineers firmly in control.

Links

Talk page with full transcript

Watch the talk on YouTube

Search this talk's transcript

Cite this talk

CDFAM Computational Design Symposium — Tokyo 2026

Matt Sederberg; Dr. Michael Scott · Coreform

Matt Sederberg and Dr. Michael Scott of Coreform present isogeometric analysis (IGA) as a route to simulating directly on CAD geometry, and outline a mathematical framework for running multi-agent LLM workflows in high-assurance engineering domains.

Sederberg covers the background of IGA, introduced by Tom Hughes and now the subject of more than 7,000 research papers, and its recent integration into Abaqus through a partnership with Dassault Systèmes. In one example, a brake caliper that took 20 hours to defeature and hex mesh, with critical fillets removed, was ready to solve in under a minute with full CAD geometry intact. With no meshing step, an agent can change the design, rerun the analysis and read the results without a human repairing the mesh in between.

Scott then previews Coreform AI, built on four concepts:

IGA as a robust physics solver for agent-driven workflows

A data weave, an algebraic relational representation that lets agents query all simulation pipeline data at roughly constant token cost, regardless of data size

Support for smaller, controllable and air-gapped models, demonstrated with Claude Haiku reaching the same result as larger models

Agent coordination posed as an energy minimization problem using game theory, with market mechanisms that penalize agents for false completion claims

He also describes formally verifying the underlying software in Lean, with more than 6,000 theorems backing returned results.

Links

Talk page with full transcript

Watch the talk on YouTube

Search this talk's transcript

Cite this talk

CDFAM Computational Design Symposium — Tokyo 2026

David Burpee · MorphXgen

Many Additive footwear platforms promise a level of customization beyond traditional footwear manufacturing processes, but how custom is it really? This presentation will explore a variety of strategies for the integration of various custom data into bespoke additive-ready algorithms in footwear design. The ability to create force-responsive structures through various types of micro lattice to tune flexibility, gradient behavior, anisotropy all within the confines of additive manufacturability parameters. Approaches include multi-agent algorithms with intelligent emergent behavior, simulation-driven anisotropic stochastic lattice structures, and body-responsive data mapping for truly one-of-one additive footwear creations.

Links

Talk page with full transcript

Watch the talk on YouTube

Search this talk's transcript

Cite this talk

CDFAM Computational Design Symposium — Tokyo 2026

Sarat Babu · SARAT

Computational design approaches to optimization are often constrained by the tools available and the established methods for generating optimal forms. With AI coding tools, the barrier for designers to develop their own software and pipelines has dropped considerably.

Sub-Optimal reframes optimization around two priorities: preserving the original designer’s macro intent, and using geometry to address the anisotropic surface roughness of additive manufacturing. The result is neither lattice application nor topology optimization, but a grown internal structure driven by simulation, surface texture and build orientation that remains largely invisible at the macro level.

The talk covers the trust gap senior engineers face with optimized DfAM parts, the use of surface texture to mask layering defects and build orientation, and how the pipeline was built in six weeks for about $400 in AI tokens using open-source libraries including geometry3Sharp and OpenVDB, with 3MF as the output format.

Links

Talk page with full transcript

Watch the talk on YouTube

Search this talk's transcript

Cite this talk

CDFAM Computational Design Symposium — Tokyo 2026

This episode is the opening keynote from CDFAM Tokyo 2026, presented by Guido Cossu and Ivo Timóteo of Braid.

Braid was founded in Tokyo in 2020 by a team drawn from physics, applied mathematics, computational geometry, large-scale simulation and machine learning. At the time, the prevailing view in AI held that scaling end-to-end learning on ever-larger datasets would solve most problems. Braid took a different position: engineering design, where physical accuracy, safety and manufacturability are non-negotiable, requires reasoning from the laws of physics and the constraints of manufacturing rather than pattern-matching from prior examples.

Their system, AURA, combines learning with formal reasoning to move from engineering intent to manufacturable geometry, replacing the manual draw, simulate, review and redraw loop with structured exploration of alternatives that the engineer evaluates and selects from.

In this talk, Cossu and Timóteo explain the semantic architecture behind that approach and show results from deployments in Japanese manufacturing, across several mass-production processes with little in common.

The Best Design Is The Design You Can Make

Guido Cossu; Ivo Timóteo · Braid

Across computational design, the part that performs and the part that can be built are rarely the same part: optimization and generative methods meet the functional targets, and the work of making the result manufacturable happens afterward, by hand, often negating the performance that was gained. Braid closes that gap differently. Rather than optimizing a seed geometry or generating candidates from data, the system designs de novo — geometry reasoned directly from what an engineer specifies: function, physical conditions, the manufacturing process the part must suit, the cost it must meet.

The mechanism is semantic reasoning rather than data-driven generation: a semantic architecture encodes physics, geometry, and manufacturing constraints with their relationships, and enforces constraint satisfaction by construction. Because the manufacturing process is reasoned from rather than filtered for, the geometry is manufacturable from the first output, not repaired into manufacturability afterward.

Braid demonstrate this on real engineering components across several conventional mass-manufacturing processes, each with its own design constraints. The examples come from deployments in Japanese manufacturing industries where tolerances are tight and errors are costly. One system handling processes with so little in common is the crucial point: each result is reasoned from principle, not retrieved from examples of similar parts — which is, in the end, the difference between a design system you can trust on a novel part and one you can only trust on a familiar one.

Links

Talk page with full transcript

Watch the talk on YouTube

Interview with Braid

Cite this talk

CDFAM Computational Design Symposium — DC 2026

Giorgia Cannici · Virginia Tech

What if material fabrication could shift from assembly to growth—and be directed with precision through external fields?

This work introduces magnetotropic plants: genetically engineered organisms in which gravity-sensing organelles (statoliths) are rendered magnetically responsive. By replacing gravitational cues with externally applied magnetic fields, plant growth direction can be actively controlled in real time. This enables programmable morphogenesis, where biological growth becomes a steerable process rather than a fixed outcome of genetics and environment.

The presentation will outline the biological mechanism, the experimental framework, and the implications of this approach for material production.

Magnetic fields act as an invisible, non-contact control layer, allowing spatial and temporal guidance of growth without mechanical intervention.

Beyond applications in microgravity environments such as space, this work suggests a broader shift in how we produce materials—moving from extractive, energy-intensive processes toward growth-driven fabrication, where form emerges from the interaction between engineered biology and designed environmental conditions.

Links

Talk page with full transcript

Watch the talk on YouTube

Search this talk's transcript

Cite this talk