Made in AI

Just for a moment, remember your old biology classes and put yourself in this situation: a species is engaged in a struggle for its survival. Small mutations in its physiology have made it more resilient to its environment. However, it has been at the cost of other qualities that, finally, end up making it disappear by the hand of the well-known natural selection.

This could be the description of how an animal, now extinct, failed to adapt and survive. But the truth is that all this is happening in a computer hundreds of times a minute and the protagonist of this story is not a mammoth but a racing bicycle chassis.

In this case, the environment is not a cold Siberian forest but a series of boundary conditions including forces, pressures, mass or temperatures that the chassis must be able to withstand. If not within an acceptable range, the chassis would not fulfil its function and would be discarded to make room for better versions.

Roughly speaking, this is how a genetic algorithm works, which is one of the key parts of the topic of this article: Generative Design.

A couple of weeks ago I attended a conference organized by Autodesk on the role of Generative Design in the future of sports. I found the content of that event fascinating and, together with the knowledge on the topic I acquired during my master’s degree, I was encouraged to write about it.

Generative Design is a new branch of design that uses today’s enormous computational power to design parts, materials and even entire products through an AI that generates thousands of possible designs that meet the constraints that have been imposed, giving the designer or engineer a huge catalogue of options for a particular problem.

This tool changes the paradigm for today’s design process for virtually anything from bicycles to plane’s wings. However, harnessing its potential is not easy. The most important part of using Generative design is the correct identification of the problem or need for which the product is being designed and the parameterization of these mathematically. This series of conditions, now written in mathematical language, is what in genetic algorithms is known as the “Fitness Function”, which is similar to natural selection. If this function is fulfilled, the design moves on to the next round, and if not, it dies out.

But how exactly does this step-by-step tool work? Well, it consists of several actions:

1. Parameterization of the problem: to find the control variables that will mutate on each iteration
2. Data collection: to be able to generate the criteria that will evaluate the designs, i.e. the Fitness Function
3. Population generation (random process under certain conditions)
4. Evaluation
5. Evolution (mutation) of the models through recombining the most promising designs. [2] Then return to step 3 until satisfactory results.

The process is refined in each iteration and finally, a set of designs that meet the fit function is obtained. The more accurate and efficient the fit function is, the better the resulting designs will be. These designs may seem like science fiction due to their organic geometries (curious, isn’t it?), but they are tremendously efficient. Needless to say, additive manufacturing is a good friend of generative design. Although, at the same time, this quest for efficiency often leads to higher manufacturing costs.

As it turns out, generative design acts somewhat as nature does. Random changes are made in the living beings that are born and the evolutionary lines that are best adapted to their environment are the ones that manage to pass their genes to the next generation. It is an iterative and to some extent “dumb” process, but this repetition over millions of years has succeeded in generating intelligent organisms, others that photosynthesize and even some that survive space travel.

Computation allows us to put this process into “fast motion” and apply it to the design of things that are easier to parameterize, model and simulate. And all this for just one part of a bike! I find it fascinating.

Success case 1: DECATHLON

Decathlon is working on a line of bicycles generated using Generative Design to create a customizable and more sustainable product. The project comes in response to market demand for stronger, lighter, more sustainably manufactured bicycles with differentiating designs. [1]

DECATHLON Bicycle Project using Generative Design — Courtesy of Decathlon.

Fork model — Courtesy of Decathlon.

Success case 2: JPL and Autodesk

Autodesk and JPL joined forces to develop the next generation interplanetary landing system using Generative Design. The final design reduces mass by 35%, [4] which is tremendously useful in the aerospace sector.

Detail of the lander prototype developed by Autodesk and JPL — Credits to Autodesk and JPL

Success case 3: Phillipe Starck for Kartell

Using the software provided by Autodesk, Phillipe Stark presented a chair that optimizes the mass it needs to fulfil its function thanks to an AI. It was presented at Milan Design Week 2019 and what I find really relevant to use it as an example is that it is the first product generated by an AI to be mass-produced. [5] In fact, you can buy it on Kartell’s website!

A.I. — Kartell

I believe this technology has the potential to revolutionize many industries and could be used as another tool in product ideation and iteration processes. This tool is not intended to replace the figure of the engineer or designer but to support it with another tool. In this way, the engineer would dedicate more time to more innovative tasks. [3]

Final thoughts

After this year of learning about it, some ideas and questions come to my mind that I will leave for the end of the article:

• In the same way that we parameterize, for example, the forces to which a part would be subjected… Could we also parameterize the aesthetics of the product? Could we program a series of Fitness Functions that allow us to generate products whose aesthetics are in tune with the visual identity of a brand?
• Deep Learning could accelerate the adoption of this technology, but how do we get data from physical products? Will we fill the physical products of the future with sensors to collect efficiency or usability data? If so, we’ll need to add data privacy policies to a bicycle.

Some links on the subject

Generative Design for Manufacturing With Fusion 360 | Autodesk

Decathlon Reimagines Lighter, Stronger, More Sustainable Bicycle Using Autodesk Generative Design |…

NASA - Next Generation of Interplanetary Lander

References

[1] Autodesk. 2020. Decathlon Reimagines Lighter, Stronger, More Sustainable Bicycle Using Autodesk Generative Design. https://www.autodeskjournal.com/decathlon-transforma-bicicletas-generative-design/

[2] Danil Nagy. 2017. Evolving Design. https://medium.com/generative-design/evolving-design-b0941a17b759

[3] Formlabs. Generative Design 101. https://formlabs.com/blog/generative-design/

[4] Autodesk. Designing a Better Lander with Generative Design. https://www.autodesk.com/campaigns/generative-design/lander

[5] Sebastian Jordahn. 2019. Philippe Starck, Kartell and Autodesk unveil “world’s first production chair designed with artificial intelligence”. https://www.dezeen.com/2019/04/11/ai-chair-philippe-starck-kartell-autodesk-artificial-intelligence-video/