Written by Audrey Woods
Imagine the process of designing, for example, a new tool. One might picture a room full of engineers brainstorming and expanding on blueprints of previous iterations. Perhaps they run through a hundred or even a thousand potential design ideas, but they can only build and test a handful of them. Each prototype requires energy and effort, and the more creative the solution, the more difficult it might be to build. Therefore, new models are limited by what has come before, not to mention the time and energy of the human minds at work. For most of history, this slow process was the only option for creating anything new.
However, advancements in computer technology are rapidly transforming the world of design, offering a way to iterate that is faster, more efficient, and works at a dramatically larger scale. Computers can run through exponentially more options, studying them in huge numbers via simulation and exploring unconventional or unusual designs that humans might not think of with their limited time, experience, and equipment. Leveraging the power of algorithms, computer scientists can improve a wide array of fields, offering robots optimized for a given application, better materials for manufacturing, personalized medical devices, etc.
As leader of the Algorithmic Design Group at CSAIL, Assistant Professor Mina Konaković Luković is working on a variety of design problems, developing algorithms which can search for better performing designs, discover designs impossible to create manually, and generally enhance the design possibilities for future engineers, artists, healthcare providers, and more.
FINDING HER INTEREST
When Professor Luković was first deciding what to study in undergrad, she had three main interests: medicine, architecture, and mathematics. She decided to study mathematics with a focus on computer science since this allowed her to, as she says, “touch upon all of these areas in some sense.” She went on to earn her PhD from EPFL in Lausanne, Switzerland, where she worked on geometry processing without machine learning or robotics. However, when she was offered the Schmidt Science Fellowship, which allowed her to study anywhere in the world, it came with an opportunity to pivot what she was working on. She chose to come to MIT because it “is one of the rare places where you can actually do this kind of exotic research, but also be on top of the game.” While a postdoc at MIT, she was mentored by Professor Wojciech Matusik, a world expert in graphics and computer aided design and well known for exploring unconventional and novel directions. This inspired Professor Luković to broaden her research into robotics, fabrication, and other elements of computational design. When she joined the MIT faculty as a professor in EECS and CSAIL, she created the Algorithmic Design Group to continue studying computational design and fabrication, geometry processing, physics-based simulation, and robotics.
Since she first decided to study mathematics, Professor Luković has witnessed a transformation in computer science. She says, “when I started studying CS, machine learning (ML) was a niche, almost nonexistent field of study, and it wasn’t popular in industry either. Now that has completely changed.” One thing that differentiates the Algorithmic Design Group is that they focus on physical objects or real-world applications, which offers the exciting satisfaction of researchers getting to see their work printed, molded, and/or used. However, it also means that their ML programs require a solid understanding of 3D space and physical constraints, which can be a challenge given the lack of good quality data for such applications. But there is no shortage of avenues for Professor Luković’s group to explore because, as she said in her 2023 TED talk, “Design is everywhere. It’s behind everything we need to create, no matter if it’s a physical or a virtual object. [In] this era of big AI success and powerful algorithms, wouldn’t it be better if we developed algorithms to help us determine the best performing designs?”
ALGORITHMIC DESIGN: CROSS-CUTTING APPLICATIONS
One of the areas Professor Luković’s group is focused on is robotics, specifically using deep learning to optimize robotic design. While working with Professor Matusik, Professor Luković was involved in developing RoboGrammar, which automatically generates the best robot design for a given terrain using ideas taken from natural language processing. This system can efficiently search through a huge variety of designs to build robots that can perform optimally in challenging environments. Such a program could be useful for roboticists in a range of fields, designing intelligent machines with clear parameters for very specific use cases. Since this project, Professor Luković has broadened the research to work on reconfigurable robots, which might empower robots with high performance on multiple tasks and in various environments, expanding what’s possible with computer design in robotics.
Not all of Professor Luković’s research is based in deep learning and AI. For example, her group is also focused on material design and deployable structures. Much of this work is based on the idea of programmable auxetic materials, or materials that can be printed in 2D and then inflated or expanded into a planned 3D shape. These materials rely on geometry processing, especially for more complex designs, and have widespread potential uses. Stents, for instance, must be inflated inside the vein to work, but right now only come in a straight shape, which can put unnecessary pressure on curved veins or arteries. Using Professor Luković’s theory, future stents could be designed to inflate in the shape of a given target geometry, like the arc of an artery. Another use case for this research is deployable architecture with materials restrictions, such as with space exploration. In this example, an inflatable habitat could be printed such that, when filled with atmosphere on a planet like Mars, it would inflate to a useful shape and structure.
When summarizing the wide-reaching potential of her research, Professor Luković says, “We have ongoing work on design of deployable structures, metamaterial design, reconfigurable robots, drug discovery, generative AI of 3D functional objects, and data-efficient ML. The applications are many, including medical devices, architectural designs, protective gear, new robotic designs, space exploration, and manufacturing.” One exciting publication that recently came out of her group was a project exploring the boundary of design with unknown physical constraints, which could further extend the potential of algorithmic design. Professor Luković explains, “When we are exploring a new design problem or a new design space, we often have no understanding up front of which designs are feasible and which are infeasible in the physical world… In our recent work we propose a new algorithm for sample-efficient search for optimal solutions while these limitations on what are feasible and what are infeasible designs are unknown. Our algorithm quickly discovers and models these limitations.”
FUTURE WORK & THOUGHTS FOR INDUSTRY
As the fields of geometry processing and algorithmic design continue to develop, the change brings new challenges, opportunities, and applications. For example, Professor Luković highlights how 3D geometry processing might be useful in modeling social media networks, creating better architectures for neural networks, or discovering new drugs or methods of drug delivery. AI is also slowly becoming more popular in geometry processing, which will further enable novel solutions.
As previously stated, one big challenge in Professor Luković’s research is the shortage of training data when it comes to the 3D world and physical limitations. There is plenty of data when it comes to images, video, and written content, but significantly less available when it comes to training machines to interact with—and create designs for—the real world. However, many researchers are working to address this shortage, several of them here at CSAIL.
Another challenge that Professor Luković faces in her work is the “big gap between research advances and industry practices when it comes to computational fabrication.” Many of these projects could have revolutionary impact in industry, but making that leap can be difficult, especially in established fields like manufacturing or slow-moving areas like medicine. But Professor Luković says, “There are many open questions and many opportunities for cross-collaboration between academia and industry. Only with joint forces will we be able to make a significant impact.”
Learn more about Professor Luković on her website or CSAIL page.