Article reprinted from: Geek Park

Original source: Geek Park

Image source: Generated by Unbounded AI

This year, 2023, will likely be remembered as the “year one” of generative AI.

AI models represented by ChatGPT have opened the prelude to AGI general artificial intelligence. AI has become an important driving force for global economic growth and has injected new "power" into the intelligence of various industries.

ABI Research predicts that by 2033, generative AI will add $10.5 billion to global manufacturing revenue.

In fact, there is no need to wait 10 years. There are already practitioners using AIGC to help them solve industrial design and even manufacturing problems.

For example, a foreign industrial designer named DIDEM GÜRDÜR used AI tools to help himself design a "jellyfish robot"; sports car company Aston Martin and even NASA have also used AI to produce corresponding products.

How did they build these things with AIGC? How long will it take for AI to completely replace engineers and designers?

01 AI built a jellyfish robot

How does generative AI “help” robot manufacturing? Engineering designer DIDEM GÜRDÜR’s “AI attempt” is a good example.

Last year, GÜRDÜR planned to build such a "jellyfish robot" in the laboratory, envisioning that it could collect microplastics from the ocean and become part of the marine ecosystem.

GÜRDÜR’s group develops robots for a wide range of “cyber-physical systems,” including smart homes and self-driving cars, that rely on the integration of physical and computing components, often with feedback loops between them.

Traditionally, researchers design cyber-physical systems through an iterative process that includes brainstorming, sketching, computer modeling, simulation, prototyping, and testing.

This is a time-consuming process that requires designers and engineers to work together and think creatively, while also refining the system's physical characteristics and control systems through extensive testing.

So, GÜRDÜR began to try a combination of the generative AI tools Midjourney and Stable Diffusion, seeking "inspiration" and trying to achieve a more efficient iteration cycle.

The initial “attempt” was unsuccessful because the keywords she chose were not “specific” enough and she did not provide any information about style, context, or detailed requirements.

For example, in GÜRDÜR’s early attempts to generate the jellyfish robot image 1, she used this prompt: underwater, self-reliant, mini robot, coral reef, ecosystem, hyper-reality.

By refining the prompts, she got better results.

For Image 2, she used the following prompts: Jellyfish robot, plastic, white background, and for Image 3, she used the following prompts: Futuristic jellyfish robot, high detail, lives underwater, self-sufficient, fast, inspired by nature.

Image source: IEEE

As she added specific details to the prompts, the resulting image more closely matched her vision of the jellyfish robot.

For images 4, 5, and 6, her prompts included: self-sufficient, futuristic electric jellyfish robot living on the seafloor, water or elastic glass-like material, shape-changer, technical design, perspective industrial design, copy-style, high detail like a movie, super detailed, moody coloring, white background.

Image source: IEEE

In this regard, GÜRDÜR advises that a good prompt must be specific and can cover many attributes, including subject matter, medium, environment, color and even mood. "If you want to include something specific in the result, then you must write it in the prompt, and you must clearly state any background or details that are important to you. You can also explain the composition of the image in the prompt, which is very helpful for design engineering products." "But if you want a certain attribute to surprise you, then you can leave it out."

She then experimented with different textures and materials until she was happy with a few designs.

Afterwards, GÜRDÜR and his team reviewed several relatively "reliable" AI designs to determine whether they could provide reference for the development of actual prototypes.

They discussed which aesthetic and functional elements would translate well to physical models: for example, the curved umbrella-like tops in many of the images could provide suggestions for the selection of materials for the robot's protective shell, and the flowing tentacles could provide design clues for flexible manipulators that interact with the marine environment.

In the process, GÜRDÜR discovered that “even if the images themselves were unfeasible designs, they could prompt us to imagine new directions we might not have considered otherwise.”

In addition, they also gained "inspiration" from the different materials and compositions generated by AI, as well as the abstract art style, and conducted more creative thinking on the overall appearance and movement of the robot. In this regard, GÜRDÜR admitted, "AI has changed the design, and perhaps also changed my thinking."

Although they ultimately decided not to directly copy any AI designs, they still generously acknowledged that "AI art is of great value for inspiration, as well as in-depth research and exploration. Specifically, AI is very useful for exploring, inspiring, and quickly producing illustrations to share with colleagues in brainstorming sessions."

In fact, in addition to providing inspiration, generative AI can also help robotics research make progress through other innovative methods, including narrowing the gap between simulated environments and the real world, promoting effective communication between robots and humans, and creating better reward models.

Although "co-creation" with AI will bring many "surprises", it also requires some perseverance.

For example, within just a few minutes of the initial prompt, GÜRDÜR saw the results produced by the AI, but she then spent hours making revisions, reiterating concepts, trying new prompts, and combining successful elements into a "complete" design.

Moreover, if you are not looking for “surprises” but for “specific” results, these AI tools can become difficult to manage. After all, humans have little control over the “iterations” generated by AI, and the results are unpredictable.

For example, when GÜRDÜR tried to change the jellyfish into an octopus, it failed miserably.

She entered the prompt words: futuristic electronic octopus robot, technical design, perspective industrial design, copic style, high detail like a movie, emotional coloring, white background, but got a "weird" image of an octopus-like robot.

Image source: IEEE

This puzzled her because jellyfish and octopuses look very similar, so why did the AI ​​generator design for the jellyfish turn out well, but the octopus design was stiff, alien-like, and anatomically incorrect?

This is probably because AI only follows the "patterns" it recognizes from the training data, and behind the formation of "patterns" is the "black box" of AI.

Therefore, AI image generators have the potential to amplify demographic and other biases in the training data, and the generated content may spread misinformation, violate privacy and intellectual property rights, and raise serious ethical issues.

GÜRDÜR is optimistic about this. She believes that "in the future, we will see some AI tools that can achieve predictability under clear constraints. More importantly, I look forward to seeing image generators integrated with many engineering tools and seeing people use the data generated by these tools for training."

02 From sports cars to NASA, AI is invading everything

In addition to robotics, generative AI is also assisting industrial design in other areas.

For example, in Aston Martin’s DBR22 concept car, designers relied on AI tools integrated into Divergent Technologies’ digital 3D software to explore the car’s form and optimize the shape and layout of the rear subframe components, thereby designing a rear subframe with an organic skeletal appearance, and then producing the actual parts of the car through rapid prototyping technology.

In this regard, Aston Martin revealed that "this approach significantly reduces the weight of the component while maintaining its rigidity. The company plans to use the same design and manufacturing process in its upcoming small-volume models."

Aston Martin uses AI to design parts for its DBR22 concept car | Aston Martin official website

In addition, other examples of AI-assisted industrial design can be found in NASA's space hardware, including planetary instruments, space telescopes, and Mars sample return missions.

NASA engineer Ryan McClelland also publicly stated that the new designs generated by AI may look a bit alien and weird, but they can withstand higher structural loads and are lighter than traditional components. In addition, they require a very short design time compared to traditional components.

McClelland even calls these new AI designs "evolutionary structures," which actually refers to how AI software iterates through design mutations and converges on high-performance designs.

NASA research engineer Ryan McClelland calls these 3D-printed parts designed using commercial AI software "evolved structures" | NASA

It’s undeniably tempting to start implementing before sufficient exploration has taken place, and even far-fetched or unrealistic AI-generated concepts can benefit from serving as rough “prototypes” for early-stage engineering design.

In this regard, Tim Brown, CEO of the design company IDEO, said, "This prototype both slows us down and speeds us up. By taking the time to prototype our ideas, we can avoid costly mistakes, such as becoming too complicated too early and sticking to a weak idea for a long time."

The practices of these cutting-edge designers and engineers have proved that, at least for now, AI cannot completely replace people’s work, but it undoubtedly provides an excellent creative tool for creative people, allowing designers to “brainstorm” more efficiently and speed up the work process.

Now the question is, after reading this article, what would you like to design using AI?

Now they are being used for increasingly complex tasks beyond what was originally envisioned. In the field of industrial design, by harnessing the power of AI, engineers can begin to think differently, see connections more clearly, consider future impacts, and design innovative and sustainable solutions that improve the lives of people around the world.