20170908-141802_edit.jpg

Cheap, simple, scalable, adaptable, and high-performance in swarms, insect-like robots have incredible potential.
Dave Burbank
Dave Burbank

20170908-144524_edit_flip.jpg

About cleverly designing robot swarms, “There are tens of thousands of bees in a typical colony and not a single one telling them what to do. They’re still able to effectively allocate workers to different tasks and to respond to sudden needs."
Dave Burbank
Dave Burbank

Peterson Collage_edit.jpg

The Collective Embodied Intelligence Lab develops algorithms to enable complex emergent swarm behavior, designs new mechanisms for simpler and more robust robots, and optimizes the performance of robot swarms.
Beatrice Jin; Dave Burbank
Beatrice Jin; Dave Burbank

20170908-134753_edit.jpg

On exploring new materials such as popcorn to enable stiffness change in soft robots, “You can imagine having a robot that walks along and decides to become part of a structure—it pops the internal kernels, becoming bigger and more rigid."
Dave Burbank
Dave Burbank

20170908-144129_edit.jpg

Speaking of the lab’s work at Cornell’s vineyards, “The idea is to have a little robot sitting in the field, and it’ll go up and feel the grapes for their shape and ripeness, estimate the yield, and potentially check for fungal infections.”
Dave Burbank
Dave Burbank

Robots, Like Insects

by Caitlin Hayes

Popular robots in television and movies often appear human-like. Even R2-D2 had a kind of humanoid face and body with a childlike bulk. In real life a human-like robot, in addition to proving difficult to program and expensive to build, also may not be the best model for certain tasks. Kirstin H. Petersen, Electrical and Computer Engineering, asks, what if some of the robots of the future looked and behaved more like a swarm of termites or bees?

The potential benefits to small-robot swarms are many. “With a lot of robots, they can work in parallel, which means they might work more efficiently and with greater adaptability than a single robot. Also, if half the robots assigned to the same task fail, the rest can still complete the task,” Petersen says. “If they’re very cheap and simple, then not only will individual robots have less breakable parts, but the collective will also scale better.”

Petersen imagines a future where, instead of sending a Spirit or Opportunity worth millions to Mars, scientists would send 100 or 1,000 small, inexpensive robots that, like bees, could operate without the direction of a central command. “There are tens of thousands of bees in a typical colony and not a single one telling them what to do,” Petersen says. “They’re still able to effectively allocate workers to different tasks and to respond to sudden needs. We want to be able to cleverly design robots such that, even though they’re very simple, they can achieve a lot of things when working as a collective.”

Petersen stumbled into robotics when she was just in high school. Then an aspiring astronomer, she got lost on the first day of a student program at NASA and found Lemur, a six-legged robot. “I just thought it was the most awesome thing I’d ever seen,” she says. “I asked the woman working with it what she did, and she was an electronic engineer. I went home and signed up and never looked back.”

Robot Swarms, like Insect Swarms

In her PhD thesis work at Harvard University, Petersen and her collaborators used a species of African termites as inspiration for robots that could build structures. Named TERMES, the robots use specially designed bricks, working without human direction to stack the bricks into simple structures. “You can give them a picture of any simple structure you want, and if possible, they’ll automatically figure out how to do it,” Petersen says. “That was our proof-of-concept.”

“You can give them a picture of any simple structure you want, and if possible, they’ll automatically figure out how to do it. That was our proof-of-concept.”

One of the challenges in this area continues to be finding universal low-level rules for all the robots in a swarm. “I can’t write a specific program for each robot because it doesn’t scale well,” Petersen says. “If I already know the program of all the robots, I can simulate what will happen. But doing the opposite, knowing the emergent swarm behavior that I would like—such as pattern formation—and then designing the program accordingly, that’s really hard, and there’s no formal understanding of how to do it yet.”

Students in Petersen's lab at Cornell, called the Collective Embodied Intelligence Lab, are now working on algorithms that could better determine the right set of rules to enable more complex tasks. A major thrust of the lab is also working to design and optimize the robots’ environment. “I like the idea of construction because we get to design the robot, the building material, and how the building grows, so we can make it happen in a way that’s optimized for the robots,” Petersen says. “In the future, we may be able to do that on a construction site or on Mars.”

This optimization of environment is inspired by nature as well. Take the species of African termites Petersen studied. “It’s amazing because millions of these centimeter-scale termites can coordinate to build meter-tall mounds. Each termite is simple relative to the mound they build; they don’t even have sight,” Petersen says. “If I took one out of the mound, it would just sit there because it lacks all sensory stimulus. But in the mound, together with all the other termites and all those tunnels it creates for itself, it’s very robust because it shaped the tunnels to fit its own limitations.”

Petersen and her team are now modifying the building materials for the TERMES robots, making bricks that fold open when they are set down. With these bricks and new sets of rules, the TERMES will be able to build more complex structures such as bridges and domes.

In addition to environment, Petersen’s lab is also developing a framework, called Open Mind, that can optimize a robot-swarm’s performance. “Imagine a camera that sits on your shoulder, that looks at all the robots, talks to them and then compares what it sees with what they’re telling it they’re doing,” Petersen says. “The system then analyses the swarm over time and tries to find faulty or suboptimal robots, informing the user of how to improve the performance of the swarm. Many specialists are working on making robot swarms autonomous; with this system, a non-expert user will be able to debug the system during runtime.”

Soft Robots

While the TERMES robots were an astounding success, landing on the cover of Science in 2014 and named one of the top 10 breakthroughs in the magazine that year, Petersen has her sights on a new class of robots that would have more dynamism and versatility—soft robots. “This work is very exploratory,” she says.

The state-of-the-science is generally that soft robots, mostly made out of flexible polymers, have to be tethered to a pump in order to move. During her postdoctoral work at the Max Planck Institute, Petersen worked towards untethering the robots. “We made this new type of actuator that allowed us to use high voltage instead of bulky pumps to drastically change the shape of the soft membranes. These could eventually be used to move our robots,” Petersen says.

Now, Petersen is continuing to explore new materials that could enable soft robots’ movement, the most surprising of which is popcorn. “So random,” Petersen says. “But in popcorn, you have a lot of energy stored up, so we can have a ton of popcorn kernels that act kind of like a fluid, but as soon as we pop them they can become irregular and can jam together and change volume. You can imagine having a robot that walks along and decides to become part of a structure—it pops the internal kernels, becoming bigger and more rigid.”

Petersen’s group is also working with the labs of Amit Lal, Electrical and Computer Engineering, and Robert Shepherd, Mechanical and Aerospace Engineering, to improve the sensors on soft robots. As one test bed, Petersen’s lab, with Justine Vanden Heuvel, School of Integrative Plant Science, will use Cornell’s vineyards. “The idea is to have a little robot sitting in the field, and it’ll go up and feel the grapes for their shape and ripeness, estimate the yield, and potentially check for fungal infections,” Petersen says. “Soft robots would be ideal for these kinds of sensitive tasks.”

The Exhilaration of Radical Collaboration

While her work continues on the electromechanical side of things, Petersen is still probing the natural world for more insights she can apply to robotics as well as ways to contribute to biological research.

She’s working with Thomas Seeley’s group, Neurobiology and Behavior, to track the pathways of bees as a tool to map and monitor the agricultural landscape. Eventually, her lab, along with Alyosha C. Molnar’s group, Electrical and Computer Engineering, would like to put tiny backpacks on bees to monitor their flight patterns. Petersen’s team is also examining how bees construct their hives. “It’s been awesome working with Tom's lab,” Petersen says. “Their work has been so inspirational to robotics, and now I’m here, and I can see them and talk to them every day.”

Like entering the robotics field, Petersen’s decision to come to Cornell was intuitive—she says Ithaca’s access to nature, which is so integral to her work, struck her immediately. She was also impressed with the quality-to-ego ratio. “People here are so down to earth, and the quality of the work they do is still amazing,” she explains.

As a systems designer, Petersen brings with her the ethic of radical collaboration. “I kind of tie a lot of people together,” she says, “and one of the things I find most rewarding is when I can stir things up—when I can get someone, anyone, excited about a different approach to science.”