Will robots ever adapt to their environment?
ASU researcher Jiefeng Sun awarded NSF grant to advance AI-powered, shape-morphing robotic systems
Jiefeng Sun, an assistant professor of aerospace and mechanical engineering in the School for Engineering of Matter, Transport and Energy, part of the Ira A. Fulton Schools of Engineering at Arizona State University (pictured in his campus lab), has received a $375,000 National Science Foundation grant for research that could lead to products that sense their surroundings and adapt their form. Photographer: Erika Gronek/ASU
The systems we rely on daily are, in many ways, still dumb.
We have cars that can reach 60 miles per hour in under 2.5 seconds and rockets with reusable boosters. However, the products we use every day, such as furniture and kitchen utensils, watches and other wearables, still share one problem: once built, their form is fixed.
What if that wasn’t so? What if a car could adjust its height to clear an obstacle, or an airplane wing could change its shape mid-flight to adapt to turbulence?
Arizona State University researcher Jiefeng Sun is building the technology that might make that future a reality.
“We want to build useful and new adaptive tools for people,” he says.
Sun is an assistant professor of aerospace and mechanical engineering in the School for Engineering of Matter, Transport and Energy, part of the Ira A. Fulton Schools of Engineering at ASU. Funded by a $375,000 grant from the National Science Foundation, he aims to develop robotic systems that can sense their surroundings and morph their shapes to achieve specific outcomes.
One might assume that Sun’s ideas come from science fiction books, but he says his inspiration comes from a place that’s familiar to everyone yet often overlooked.
“I draw inspiration from nature,” he says. “A human hand, for example, is controlled by roughly 27 muscles that coordinate perfectly to grasp a cup or make a fist. My team studies such natural systems to design robotic structures with similar capabilities.”
Sun has already successfully figured out how to develop low-dimensional robotic systems with unprecedented adaptability. In this project, he’s embarking on a journey to expand his previous innovations to extremely complex and robust systems. While that’s exciting, he says the road ahead could be bumpy.
“Controlling a single module is easy, but controlling an assembly of the modules to achieve high-dimensional shape change is difficult,” he says.
The challenges Sun needs to overcome are twofold.
Building a robotic system that can change its shape in response to its environment requires specialized hardware. Just like a sea turtle needs a flat shell to glide through water, while a land turtle’s arched back enables it to walk on rough terrain, Sun says a robot’s hardware influences what it can do.
“If we can design artificial muscle that can morph from one shape to another, we can create complex robots with entirely new capabilities,” he says.
Traditionally, robots are built using rigid actuators, which limit how smoothly they can move or change form. Sun developed fiber-reinforced pneumatic-driven artificial muscles that behave like a biological tissue. The small, flexible actuators contract and expand using air pressure rather than heat, allowing the robot’s muscles to respond almost instantly.
The discovery of these new actuators gives Sun confidence to solve the second part of the puzzle.
“With this new artificial muscle, which is much more powerful than previous muscles, we can achieve more complex shape changing,” he says.
Making robust robots autonomous
Sun’s next mission is to use artificial intelligence, or AI, to make complex robotic systems autonomous.
“We’re capable of building complex systems using our artificial muscles. We just need to figure out how to control such a robust system in real time,” he says.
He likens his research to a toddler learning to stand.
“Before kids can stand up, they learn how to control the foot muscle, the belly muscle, then the neck muscle, and all the muscles involved in the motion,” Sun says. “Eventually, they put it all together to coordinate standing.”
Similarly, because the systems Sun aims to develop are extremely complex, he plans to use the Koopman operator, a mathematical tool, combined with a graph neural network to map high-dimensional robotic systems to simple systems he already knows how to control.
By breaking a large system down to its smallest units — like a kid learning how to walk — Sun will need only to figure out how to coordinate the units to achieve a certain goal. His team plans to train AI models and create algorithms to recognize patterns and control how a system’s shape changes autonomously.
Taking a system with a thousand components as an example, Sun says AI will be able to predict which of the parts need to change shape for the whole system to adapt to a new environment.
“We want the robot to change its shape instantly,” Sun says. “That means we need AI models that can process huge amounts of data and advanced sensors to notice a change in the environment.”
To him, this project couldn’t be more timely.
The recent breakthroughs in generative AI and machine learning, spearheaded by technology companies such as OpenAI and Google, have opened doors for his research to reach new heights.
Sun emphasizes something else that will be critical to the success of his work.
“At ASU, we have many resources we can leverage,” he says. “People are working on different aspects of robotics in various applications, like health care. It’s really easy to find collaborators and people with different expertise.”
When he projects into the future, Sun gets excited about the potential impact of his work.
“My ultimate goal is to combine the novelty of our artificial muscle and AI to make more intelligent and useful products,” he says.
