Original Source Here
Algorithmic Creativity with Inspiration Driven Design Process
Creativity is commonly understood as an ability to create new concepts that are non-intuitive, unobvious, and beyond the imagination of others. It is often found that creative people have exceptional ability to see things from multiple perspectives and connect concepts from diverse domains during design; be it a product, artwork, process, or any solution.
This method is so powerful and sought after, that often designers are recommended to imbibe this style or are trained to practice what is commonly referred as “generating ideas by drawing analogies”. Here we delve deeper into the practice of generating ideas by drawing analogies and present how this can be used in achieving algorithmic creativity; aka creativity achieved computationally.
Analogy driven design often starts with an ‘inspiration’. For instance, the early versions of Ford’s assembly line were inspired from the processes in grain warehouses. To generate ideas by analogy, one establishes a connection between what one is designing with something that exists in a similar or disparate domain.
Bio-inspired design approaches, commonly referred to as biomimicry, are particularly worth mentioning here, as nature has served as a source of inspiration to designers since time immemorial.
One notable example traces back to 1480s when Leonardo da Vinci studied birds and came up with his first Ornithopter design for human flight. This design was never built but it has inspired several designers and engineers in designing gliders and later led to the breakthrough invention of airplanes by Write brothers.
Climbing pads supporting human weight have been inspired by the biomechanics of gecko feet. George de Mestral invented Velcro after noticing burrs sticking to his dog’s hair. Human eye’s vestibular ocular reflex (VOR)-has inspired to precisely move camera to minimize blurriness caused due to view point changes in camera.
In summary, to generate ideas by analogy, one establishes a connection between what one is designing with something that exists in a similar or disparate domain. In other words, one tries to look for an inspiration (source) that resembles or works like what one is trying to design (target).
Given this powerful method, we are faced with two important questions: Firstly, can we empower designers’ creativity with a set of tools that help them practice analogy driven designs effectively and efficiently? Secondly, can algorithms generate designs using an analogy driven design approach?
To answer this, let’s look at the steps a designer would follow while using the analogy driven design process:
- The first step is inspiration selection, where the designer primarily looks at inspirations and assesses if they can be useful for his/her design. Often, the search for an inspiration is driven by the design objective. Imagine that you are designing a high-speed robot especially for movement on rough terrains. In this case, you would search for an inspiration where speed is the key element. Here, a cheetah can be a good inspiration from nature.
- The next step is to identify elements that match the target design objectives. For instance, the designer would need to look for the mechanism that contributes to a cheetah’s speed and may want to assess if existing mechatronics can support these; e.g. specific legs or body movements.
- The last piece is to apply the identified elements to the target design; here the designer applies / adopts the identified elements to his design. For instance, the cheetah weighs lesser, has a body shape that minimizes air resistance, and has specialized eyes for wide viewing angle. All of these could potentially become important design elements to consider and apply in the robot design.
Empowering designer’s creativity
Let’s look at how technology can assist designers in analogy driven design process:
Inspiration search: Search tools that work on curated catalogues of inspirations can empower designers in their ability to generate ideas by quickly accessing multiple inspirations. A powerful search and similarity matching system can help them in finding inspirations and filtering out matching elements.
Design generation: with the emergence of artificial intelligence, especially generative networks, designs can be computationally generated by applying inspiration to the target. Neural style transfer algorithms have shown good success in applying elements from one artwork to another. Figure 3 shows computationally generated artwork ‘B’ which is obtained after applying style artwork (thumbnail image) on source artwork ‘A’. Similar attempts have been made for 3D models also; however, they are still at nascent stage. For instance, Figure 4 (a) shows an example of applying 2D pattern on 3D models. Figure 4(b) shows an example of analogy driven 3D style transfer. Here, first an analogy relationship is established between source armchair and target sofa (top row and applied on exemplar armchair to generate the output. This is achieved by a series of geometric manipulations.
Design generation with 3D deep neural networks
Most of the successful 3D model generation approaches rely on geometric methods which are often computationally expensive. Given the success of 2D style transfer, attempts are also being made at applying neural style transfer on 3D models but this has had limited success.
With an ambition to achieve algorithmic creativity for 3D modeling, we have developed multiple approaches to achieve style transfer of 3D models. Our approaches have shown significantly better outputs than the existing solutions. Without delving deep into the technical details, which will be the topic of our next blog, our approach uses deep neural networks for 3D point clouds and meshes to achieve style transfer
To give a concrete example, imagine you are creating a 3D model of a chair, and your goal is to give the chair an antique look. The inspiration here could be design elements from ancient architecture. Next, you need to segment your base chair model into different parts. After the segmentation, you need to find antique/ ancient architectural elements which are compatible with the chair parts. For instance, you may find an architectural pillar structure that would go well with the chair legs and an antique mirror frame as the inspiration for the back of the chair (Figure 5). These elements now can be adopted in the base chair design. All the steps described here, such as searching matching elements from a catalog of inspiring elements, 3D model segmentation, similarity matching, and adopting 3D model can be achieved with advanced AI algorithms for 3D models. This example also demonstrates that AI algorithms can indeed generate creative designs without human intervention.
Computational and AI based tools have shown great potential to empower designers during the most creative phases of design. These tools are also getting better in their creativity with advancements in AI algorithms. In fact, when it comes to only form factor, the existing solutions do a decent job at automatically generating multiple designs. However, there is a large scope of improvement in the existing AI based 3D generation algorithms and a potential to take this algorithmic creativity to the next level, i.e. from shape or form to include functionality too. That will truly define the next generation of algorithmic creativity.
1. Ornithopter https://en.wikipedia.org/wiki/Ornithopter
2. Learn How to Use the Best Ideation Methods: Analogies https://www.interaction-design.org/literature/article/learn-how-to-use-the-best-ideation-methods-analogies
3. Analogies as Creative Inspiration Sources in the Design Studio: The Teamwork https://www.atiner.gr/journals/architecture/2015-1-1-4-Casakin.pdf
4. Generate creative options with “forced analogy” https://www.pallikkutam.com/edu-landscape/generate-creative-options-with-forced-analogy
5. Analogy based idea generation with TRIZ https://triz-journal.com/analogy-based-idea-generation-with-triz/
6. Analogy-Driven 3D Style Transfer. https://people.cs.umass.edu/~kalo/papers/ShapeSynthesis_Analogies/2014_st_preprint.pdf
7. A 3D shape generative method for aesthetic product design https://www.sciencedirect.com/science/article/abs/pii/S0142694X19300791?via%3Dihub
8. Image Style Transfer Using Convolutional Neural Networks https://www.cv-foundation.org/openaccess/content_cvpr_2016/papers/Gatys_Image_Style_Transfer_CVPR_2016_paper.pdf
9. Neural 3D Mesh Renderer. https://arxiv.org/pdf/1711.07566.pdf
10. Inspired by Human Eye: Vestibular Ocular Reflex Based Gimbal Camera Movement to Minimize Viewpoint Changes https://www.researchgate.net/publication/330446149_Inspired_by_Human_Eye_Vestibular_Ocular_Reflex_Based_Gimbal_Camera_Movement_to_Minimize_Viewpoint_Changes
Trending AI/ML Article Identified & Digested via Granola by Ramsey Elbasheer; a Machine-Driven RSS Bot