Generative Design AI Workflow for Autodesk Fusion 360

The Product

An integrated cloud based platform for CAD, CAM, CAE, PCB & PDM core for designers, engineers and machinists and teams working across product development lifecycle.

But what is Generative Design AI?

Generative Design AI is a parameter based AI algorithm for optimization and generation of various 3D CAD models.

Know more about Generative Design AI 🡵 workspace in Autodesk Fusion 360 🡵

Workspace & Tool Background

This section explains knowledge about the Autodesk Fusion 360 workspaces & tools for individuals who are not privy with the platform.

'Generative Design AI' Workspace

This workspace is primarily divided into three regions and is primarily used to add & assign parameters to the Generative Study subsequently leading to generation of various solutions.

Conventional 'Design'/CAD Workspace

This workspace is primarily divided into four regions and is primarily used to develop geometries through the conventional processes of CAD.

Workflow for using Generative Design

There are various ways of using the Generative Design pipeline but this is the most conventional high level process used to generate various solutions.

Base knowledge about Generative Design Parameters

Various parameters assigned within the space have been explained here to provide more context about the subsequent research and design for the project.

Any geometry which is yellow in color indicates to the software that it has to use this geometry as the base value for further calculations.

Any geometry which is green in color indicates to the software that the geometry is to be left untouched and is supposed to remain in the 3D form it has been originally designed.

Any geometry which is red in color indicates to the software that there cannot be any geometry generated in the 3D volumes which are occupied by the red geometries.

Any physics parameters such as loads or constraints are assigned using these tools.

Context

Gen AI in the day-to-day

Gen AI excels at general purpose automation which has made these tools commonplace and accessible for a range of day to day activities from replying emails to managing your finances.

But Generative Design AI process in Fusion 360 is wayyyyyyyyy different!

A bunch of pre-requisites, a different input modality and, on the go visualization of 3D models lead to a pretty steep learning curve.

Pre-requisite Knowledge

A user needs to gather a bunch of information and should have a background knowledge in Fusion 360 & Engineering before getting into GenD AI.

Input Modalities

The user has to use tools to assign parameters which would be used by AI as data/requirements to generate models.

So ideally....

The user should have an Engineering or Design background.

But....

almost a quarter of Fusion 360 users are novices to Engineering or Design!

24%

of the total user base has Educational Accounts

Which is about

5 Million Educational Users!

Now this user base is ‼important‼ as

Autodesk's market capture strategy is based on free Educational Licenses to its software during higher education.

So for Business Context

I questioned....

Could Generative Design AI become the "day-to-day" AI in 3D CAD contexts?

and potentially become a factor to drive market capture for Autodesk in the 3D CAD software industry?

That led me to do some Free-Form Research with engineering students @ Purdue.

and I found out that....

Novice users struggle with understanding the tools & processes of Generative Design due to the steep learning curve, and relative newness of the technology.

no one wants to spend hours watching youtube videos or days reading documentation simply to learn a new tool.

Formalizing Research

Understanding a novice user's approach to the Generative Design Workflow

A structured research approach to identify and understand how novice users perceive, approach and learn to use the Generative Design workspace.

The Process

Building a theoretical foundation

Informed from free-form conversations I decided to build a strong theoretical base through literature reviews to understand learning models, cognitive theories and technology acceptance models.

User Study: Focus Groups in Engineering Classrooms

Focus group sessions with Undergraduate, Graduate & PhD students. as they emulate a real world scenario of engineering teams, enabling richer data.

Sense-making

Understanding all the data collected and what it means to identify emerging themes.

Theoretical Foundations

A structured Literature Review to understand existing research in the area, as well as reference methods.

Cognitive Load Theory

(Sweller, 1988)

Novice users experience high cognitive load due to the complexity of CAD interfaces, which can overwhelm working memory and hinder learning.

ICAP Framework

(Chi & Wylie, 2014)

Cognitive engagement in learning tools is maximized when users transition from passive (e.g., watching tutorials) to constructive or interactive modes (e.g., experimenting with parameters).

Strategic Command Aggregation

(Bhavnani et al., 1993)

Novices often use inefficient, trial-and-error command sequences (e.g., manually editing each vent in CAD), while experts leverage aggregation strategies (e.g., batch-editing with fences) to reduce errors and time

Reliance on Instructors & Heuristics

(Lee et al., 2010)

Students often default to seeking immediate instructor assistance when encountering CAD challenges, expecting educators to provide direct solutions. This reliance stems from ingrained classroom dynamics (the "didactic contract") and limits independent problem-solving.

Game inspired interfaces & usability gaps

(Kosmadoudi et al., 2013)

Traditional CAD interfaces often fail to sustain user engagement due to complex menus and passive workflows. Game elements like real-time feedback, progress tracking, and reward systems can make interactions more intuitive and motivating.

User Study: Focus Groups in Engineering Classrooms

Over multiple rounds of focus group sessions the study helped decipher the most important insights and themes which were subsequently used to design the interventions.

6 Undergraduate Students

5 Graduate Students

2 PhD Students

1 Teaching Assistant

1 Professor

Phase 1

30 mins

Introductory Presentation on Generative Design, its importance and its use cases.

Phase 2

15 mins

Case studies of how Generative Design has been implemented in the Industry.

Phase 3

60 mins

Tutorial Exercises + Introduction to Tools & Interfaces + Discussion & Observation

Phase 4

60 Mins

Discussion & QnA

Phase 5

15 mins

Conclusion & Closeout

and I began with one question....

What is a novice user's workflow expectation while using Generative Design?

Sense Making

It was a LOT of data and, deciphering, categorizing and understanding everything was crucial to the design recommendation.

6 Hours

of session recordings,

16 Session Assignments

analyzed & deciphered

Countless

observations

helped reveal

4 overarching themes

Findings

A range of insights - structured & unstructured

From overarching themes to targeted mental models regarding Generative Design workflows, everything pointed towards providing learning aide for novice users.

The overall synthesis revealed 4 overarching themes which kept repeating with multiple users over multiple sessions.

Users experienced significant cognitive barriers due to the transition from Traditional CAD workspaces to Generative Design workspaces.

There was an initial overwhelm repeatedly observed due to the newness of the tools and their steep learning curve in the Generative Design workspace.

Mental Model misalignment between Traditional CAD space & Generative Design space leads to confusion in the process of assigning parameters.

Users did not discover the 'Modeling Intent' which is a core concept required for the Generative Design workspace.

So it all started when the students....

Reflected a 'Trackback' approach in their mental models for traditional CAD workspaces

where they broke down a 3D geometry into smaller and smaller elements.

Existing Mental Model for Traditional CAD practices

and this mental model reflected....

The organization of tools in the conventional CAD a.k.a. Design Toolbar.

where the logical order of usage of tools is generally arranged in a left - right manner, due to which the existing mental model of users has developed.

But Generative Design is different!

where the conventional CAD toolbar is used to 'build/create' geometries, the Generative Design toolbar is primarily used to 'assign' parameters.

There is no inherent logical order of tools reflected to a novice user, hence users utilize their existing mental model understanding that the tools are supposed to be used from left - right.

How is it different one asks....

The concept of 'Modeling Intent' for conventional CAD processes

is what you Model is what is Manufactured.

Tools Breakdown for Generative Design Toolbar

but for Generative Design the....

Concept of 'Parameter Bodies/Geometry' is crucial.

To run a successful Generative Design AI study, a user would have to model 3D bodies/geometry which would be used as 'Parameter bodies'.

so....

The concept of 'Modeling Intent' for Generative Design AI CAD processes

is not to model to manufacture, but to create 3D bodies as a means of feeding the AI algorithm data.

So where are these concepts and differences communicated to the users?

Technically NOWHERE!

Though there is nudge/hint of these concepts pop up in the 'Edit Model' tool section inside the 'Learning panel'.

But the problem was that any 'Explicit Explanation' of the concepts was found nowhere!

This is because novice users expect explicit explanation of tools and processes they are using for the first time.

However, the novice users used 'Help Pop-ups' on hover the most to understand what a tool does.

which is a quick 'short attention-span' way to understand the function of the tool.

and I asked....

How might I clearly communicate the concepts of 'Modeling Intent' & 'Parameter Bodies' in the Generative Design Interface?

Such that the users do not have to spend time reading documentation or watching youtube videos just to understand how to correctly operate Generative Design AI.

so to understand where exactly to communicate these concepts....

I did a deeper analysis of the Generative Design Toolbar.

where I found that the toolbar has some tools directly needed for the Generative Design process, whereas some tools are for learning GenD, some are good to haves and some others used by advanced users.

Must Haves

These tools are base needs for the overall process of setting parameters for Generative Design studies.

Advanced

These tools are optional but good to have in advanced Generative Design Studies.

Good to Haves

These tools are good to have in terms of assisting the study, however not needed all the time.

Good to have tools for generative design

Redundant

These options here are usually accessible from other places in the overall software interface.

Redudant menu options for generative design

and this analysis translated into....

The ideal flow of using Generative Design Tools

This user flow is referenced to provide context about how a learned user would navigate through the interface to produce Generative Design Results.

An expert's workflow of using Generative Design

here....

The 'Edit Model' tool popped up again!

It is the only tool through which one can add/edit bodies in the Generative Design space.

The expert user would use the 'Edit Model' tool to add parameter bodies, however they have developed the knowledge through experience, leaving the novice user in confusion.

But....

The 'Edit Model' tool inherently didn't communicate to the user that they have to 'Add' bodies.

Unless the user accessed the 'Learning Panel' and spent hours learning the intended way of using Generative Design.

Which communicated the 'addition' aspect, but did not communicate if they were 'Preserve' or 'Obstacle' bodies.

And this is what confused a novice user the most!

So I came up with one....

Major Design Goal

To design a method of communicating the concept of 'Modeling Intent' more efficiently in the Generative Design Space.

That too in a way that it reduces the learning curve for a completely novice user.

Design

Updating the Generative Design Toolbar

Adopting a discovery-based approach and, integrating user feedback at each stage the solution addressed novice users' needs of not accessing learning modules at every step of the way.

To keep changes to the minimum....

The first direction was towards improving the grouping of tools and UX copy to provide more context to the novice user.

Comparison of Current 'Generative Design' Toolbar and 'Proposed Generative Design' Toolbar

However, after some testing, this solution didn't successfully communicate the concept of 'Modeling Intent'

So in order to communicate Modeling Intent & contextualize Parameter Modeling....

What if the functionality of the 'Edit Model' tool was bifurcated?

What If? Combining functionality tools in Generative Design

This method would provide users with separate 'Contextual Environments' to 'add' both 'Preserve' & 'Obstacle' geometries instead of 'Edit Model'.

Comparing V1 of design update and final version of Generative Design Toolbar

So finally....

A refined toolbar geared towards greater context so that novice users can understand modeling intent in the 'Parameter Geometry' section to navigate & utilize the Generative Design workspace with greater ease.

The final solution of GenD Workspace's Toolbar

Greater Context about 'Parameter Body' modeling through the 'Parameter Geometry' menu supporting the need to communicate modeling intent.

Compartmentalizing the parameter assignment(+add) task into 3 main areas: Parameter Geometry, Assigning Structural Conditions & Manufacturing.

Updated UX copy to provide greater context for novice users during the process.

Testing

Concluding Design & diving into deeper testing

Testing is still an ongoing process for the later versions of the design & will be updated soon!