The new robot paradigm
Published: May 1, 2018 • By Emily Adams, original post
Every day, it seems that robots are becoming more realistic-looking and capable, from Sophia, the humanoid robot that was granted citizenship in Saudi Arabia, to robots that can assist in surgeries.
But for computer scientist Nikolaus Correll, robots are still missing something—they’re falling short of the complexity and elegance of nature. When he sees ants constructing anthills or a horse running, only one question comes to his mind: “How can we engineer those kinds of things?”
Correll believes the key to mimicking nature lies in embedding materials with thousands of tiny computers that work together to sense, compute and carry out a function—all without relying heavily on a central processing “brain.”
Making materials smarter means robots would no longer look like the “clunky” things we have now, he said. In fact, they could look like almost anything at all.
In the coming years, he will lead a research effort that will bring together electrical engineers, mechanical engineers, computer and materials scientists, and others. The goal is to take these so-called multifunctional materials from proof-of-concept to prototype and determine how they can best be deployed in everything from aerospace to biomedicine.
What we can engineer is really orders of magnitude apart from the natural world in terms of complexity, in terms of elegance, in terms of multifunctionality.
Correll has already been playing in this space for several years with his swarm robotics work. With swarm robotics, the goal is to create a team of simple robots that work together to accomplish a complex task.
But it’s been tough to convince other researchers, like materials scientists, that they’re working on the same thing. For instance, many materials researchers have tried to recreate cuttlefish camouflage by combining different materials that inflate to change the pattern and texture of the skin.
While Correll has created droplet LED robots that are a type of camouflage—changing their configuration in response to an external stimulus—Correll said materials scientists often don’t see the connection. For them, “the camouflage has to be a material” because they don’t understand how to create the complex algorithms a full system requires, Correll said.
Correll says that if they can all work together to make computers small enough, they can do things like pour them into a liquid or gel medium to form a functional skin for a prosthetic limb. But to make the computers that small, you have to start by making them much more efficient.
Enter a new type of computer architecture called neuromorphic computation. When the call went out for new interdisciplinary research areas in the College of Engineering and Applied Science, electrical engineer Sean Shaheen submitted his own proposal for this topic. But he and Correll soon realized they had a similar goal—they were both trying to mimic biological systems.
Neuromorphic computing seeks to make computers that work more like the human brain. In current computing architecture, one transistor sends signals to another, but that signal is always the same strength. In a human brain, a synapse receives thousands of signals and adds them all together. If the resulting combined signal is strong enough, the neuron fires and sends the signal to the next synapse. Because it’s not sending along every signal that comes through, the system is much more energy efficient.
“The goal is to do very efficient computing at very low power,” Shaheen says. “If you look at the human brain, it does all of this amazing cognitive processing that current computer systems can’t do, and it does it with the power of about 20 watts,” about the same power as a laptop computer.
His plan is to integrate neural hardware into Correll’s robotic materials. For instance, if a manufacturing robot has an artificial hand with thousands of sensors in the fingertips, a neural system would allow it to simplify all of those inputs into a smaller amount of information for the central processor.
While today’s computers are essentially very powerful calculators, neural systems enable things like better image recognition. Shaheen envisions creating digital assistants for Alzheimer’s patients, to help with facial recognition, or building personalized learning assistants for students.
“The human brain is not the best at number-crunching,” he said. “I can’t tell you what the square root of 3,277,000 is, at least not off the top of my head. But I can pick out a picture of my mom amongst dozens of pictures within a fraction of a second.”
While Correll and Shaheen tackle their parts of the multifunctional materials equation, mechanical engineers like Christoph Keplinger will be working on designing better actuators—also inspired by nature.
Earlier this year, Keplinger’s research group announced that they had developed a new type of soft, electrically activated device capable of mimicking the expansion and contraction of natural muscles. These devices, which can be constructed from a wide range of low-cost materials, are able to self-sense their movements and self-heal from electrical damage.
Overall, Correll is driven by the intrigue of breaking down natural systems to their basic parts and functions. He points out that humans are just a collection of atoms that have physical interactions.
“This idea that all these things are made from these basic building blocks that you can just put together is, I think, very stimulating,” he said.