Engineers Agree: Nature Makes the Best Robots
In nature, elegant engineering solutions abound. The robotics world is working to unravel them.
My escorts and I walked for five solid minutes through a converted World War II–era warehouse, winding through a maze of dim corridors and a cavernous rail bay, then through a lab full of spacecraft skeletons in the midst of prototyping,. We finally reached the workbench where the Navy is building… a robot squirrel.
“Squirrel” is a bit of a stretch, as the first fully built-out version of the Meso-scale Robotic Locomotion Initiative (MeRLIn) will weigh 10 to 20 pounds when it’s finished this spring — a monster of a rodent, by anyone’s definition. The robot in its current form consists of a rectangular manifold and the 10th iteration of a dog-jointed leg, mounted on a sliding aluminum strut. A bright-blue 3-D printed model nearby showed how it will look when complete: a headless, four-legged machine about the size of a Yorkshire terrier.
But when the project’s engineers fired it up to give me a demonstration, I saw why they refer to MeRLIn as a squirrel: Despite its tiny motors and hydraulic-driven pistons, it can jump like hell.
MeRLIn is just one of the recent robots that have animals to thank for their inspiration. The animal kingdom is rife with examples of clever sensing and movement, and efficiency is king in the battery-driven, limited-power world of autonomous robotics. The ability to imitate a kangaroo’s jump, for instance, would realize an ideal tradeoff between power and performance: The tendons in these marsupials’ formidable hind limbs store energy between every stride, allowing the animals to travel long distances with relatively little energy expenditure.
Biology is behind some of the most innovative robotic designs emerging today: Look at UC Berkeley’s Salto, inspired by the high-jumping African bushbaby, or the University of Virginia’s mantabot, modeled after cownose rays of the Chesapeake Bay.
It’s easy to see why. Biologically inspired designs have clear advantages when it comes to accomplishing tasks for which the human form is poorly adapted. From tiny flies to deep-sea fish and even microbes (some fuel cells are driven by microbial chemistry), nature has tinkered and tweaked amazingly effective ways to get jobs done. Millions of years of evolution has made animals incredibly effective at the jobs they do — flying, jumping, walking, and swimming; sensing in invisible spectra; and likely more abilities we haven’t yet discovered.
But far from being mechanical replicas of animals, the bio-robots being built today are advancing the goal of distilling these elegant biological solutions. The push now is to parse what those strategies are, pare them down into their principal essences, and harness them for our own purposes. While scientists and engineers are building components that can move better, processors that can think deeper, and sensors that can detect more finely, though, stitching it all together into a truly functional, mass-producible package remains an elusive task.
Falling Before Walking
If MeRLIn looks familiar — well, it should. Glen Henshaw, the lead investigator of the project, said his team makes no bones about the fact that MeRLIn is inspired by much larger and heavier ancestors that have already found a good measure of Internet fame, including Boston Dynamics’ L3 and Big Dog and MIT’s Cheetah.
What the Navy Research Lab engineers are aiming for is a smaller, quieter, and more agile robot, one that doesn’t require two strapping young Marines to set it up to check out potential hazards. But building MeRLIn is not as simple as merely scaling down all the parts to make a robot that can fit into a soldier’s rucksack. It’s also a process of understanding how and why certain gaits function, why those gaits are appropriate for varying terrain, and how to build a robot that can learn to adapt and choose the right ones.
Arriving at MeRLIn’s bench, Controls Engineer Joe Hays inputted several test commands to a computer, making the robot’s leg twitch and jerk. After he removed its support strut, MeRLIn’s single leg held up its brick-sized body under its own power, now charged with hydraulic fluid.
Moments later, with a lightning spasm, the leg launched merRLin nearly three feet into the air, guided up and back to the table by its vertical metal rail. Repeating this exercise three more times, the robot hit the ceiling of its protective enclosure after one final, powerful jump, landing so heavily that its leg collapsed.
“There’s a lot out there we still don’t know about animal locomotion, frankly,” Henshaw said. “And we really don’t understand the neuromuscular system as well as we’d like. We’re trying to build something without knowing exactly how it should walk.”
The team is still working out a few more issues with the hydraulics but has found good success with an adaptive algorithm that probes out and corrects for uncertainties in the hardware’s circuitry at a rate of once per millisecond. They expect to have it try to jump from the ground to a desk within several months.
At the University of Pennsylvania, Avik De and Gavin Kenneally’s Minitaur is another recent super-small, lightweight quadruped, created under the guidance of Dan Koditschek. Weighing scarcely 14 pounds, their little bot has an endearing, bounding gait. Endearment quickly turns to wonder, though, when you watch videos of their creation clambering up stairs, climbing fences, and jumping to unlatch a door handle.
De and Kenneally drastically cut the bulk of their bot by using free-swinging, direct-drive legs instead of traditional gear-driven legs. The motors act as feedback sensors to the robot’s software, detecting and adjusting the torque they deliver 1,000 times every second. The result is a robot that can bound along slowly or quickly, climb stairs, and jump up and swing a set of legs around to hook a door handle to open it.
Though it’s still far from autonomous, lacking sensors and control systems that would allow it free range, Minitaur’s unique, adjustable pogo-stick action demonstrates that agility is possible even without large, powerful drive mechanisms. It’s also made from commercially available parts.
“Clearly there’s plenty of motivation for having legs, but the current state of the technology is not mature enough and prohibitively expensive,” De said, referring also to Boston Dynamics’ Atlas robot — more than capable, but proprietary and pricey, so not easily replicated. “We wanted to make a robot that was accessible to other people so they could try to implement the platform for their own applications.”
By his own admission, Howie Choset is afraid of snakes. It’s wonderfully ironic, then, that his best-known works can best be described as snakelike.
Choset, an associate professor at Carnegie Mellon University in Pittsburgh, has been working with snake robots since he was a graduate student, and he’s racked up a litany of accomplishments. He runs CMU’s Robotics Institute — a lab where many of the creations in progress feature the repeating body segments of snakes. He’s also an editor of the recently debuted Science Robotics journal and has authored a textbook on principles of robot motion.
And just to stay busy, he’s also founded two companies: Hebi Robotics and Medrobotics. The latter’s advanced endoscopic surgical tool, the Flex Robotic System, received FDA approval in 2015 for use. Though Choset is now no longer formally affiliated with Medrobotics, he said that watching a live operation in which the robot was used was the highpoint of his professional experience.
Choset demurs on whether the Flex was inspired by snakes; he said the robot’s serpentine form was designed with the twists and turns of human inner space in mind. But other, more recent work has most certainly involved looking at snakes and modeling robots after them, especially through collaboration with Georgia Tech’s Dan Goldman, a physicist whose research in biomechanics has led to the creation of robots inspired by the movement of crabs, sea turtles, cockroaches, mudskippers and sandfish.
Choset also acknowledges the influence of one of the original pioneers of bio-inspired robotics, Robert Full, who runs UC Berkeley’s Poly-Pedal lab. By studying how cockroaches move and how geckos climb vertical surfaces, Full, Choset, and others seek to boil these secrets down into general design principles that can be applied in novel ways.
“Should we copy biology? No. Ask a biologist for that,” Choset said. “What we want is to cherry-pick the best principles and go from there.”
Together, Choset and Goldman, along with Zoo Atlanta’s Joseph Mendelson, studied the movement of sidewinder snakes, ultimately characterizing their sharp-turning movements as a series of shape-shifting waves. Applying that knowledge to the programming for his robotic snakes, Choset’s team was able to make them clamber over mounds of sand, a previously impossible task. Understanding how snakes change their body shape to get themselves around has also allowed Choset to build snake robots that can writhe up posts and the insides of door lintels, something he envisions as eminently useful for exploring dangerous interiors — say, a nuclear power plant or the inaccessible confines of an archaeological site.
“I’m humbled by the fact that biology is so complex and can only hope to take a little bit of it and put it into our robots,” Choset said. “But we’re not replicating animals to the fine degree and capability that animals have. What we want is to build mechanisms and systems that have great capabilities.”
His description of his own advances and his students’ achievements and discoveries as fairly serendipitous also applies to how robots like these will emerge into the world as they mature. Slowly, in small increments, the research is getting there, he said.
“Evolution is haphazard, too,” Choset asserted. “There’s no one tipping point, only a sequence of developments that, seen from the outside, looks like a big breakthrough.”
A Critical Crossover
In the main, engineers can’t be expected to know how biology works, which makes collaborations between engineers and biologists critical. At the University of Chicago, biologist Mark Westneat’s studies of wrasses, a class of fish, led to a collaboration with the Navy, resulting in a slow-moving but agile underwater drone that can hover in place. Known as WANDA (which stands for “Wrasse-inspired Agile Near-shore Deformable-fin Automaton”), drones like these will be useful for inspections of ships’ hulls, piers and oil rigs.
High-speed photography was central to the effort nearly 20 years ago, when Westneat first started doing imaging studies of the wrasses and before the Navy got interested in the work. In a flow tank with a constant current, which Westneat calls a “treadmill for fish,” wrasses swim along happily, using only their pectoral fins to maintain a fixed position in the tank while high-speed cameras capture every detail of that movement at 1,000 frames per second.
Combined with the biologists’ highly detailed knowledge of the fish’s anatomy — how its fin rays attach to its muscles, how the nerve endings in fin membranes relay stresses and tension — the photography enables a deep knowledge of how exactly the wrasses propel themselves through the water with the twisting and torsion of their characteristic penguin-like flapping stroke. The wrasse’s ability to essentially hover in place while keeping its body still in even in strong or fluctuating currents makes it an ideal species to model for a novel type of agile underwater vehicle, said Jason Geder, a lead engineer on the WANDA project at NRL.
“Traditional propeller- or thruster-driven vehicles don’t have that kind of maneuverability or have too high of a turn radius,” Geder said. “This was a good fish to model, because if we wanted to have a rigid hull for payloads at the center of the vehicle, we could get similar performance just using this kind of pectoral fin movement.”
Westneat thinks that newer 3D photographic capability can advance the research even further. “For the fish, it’s life or death, but for us, a better understanding of efficiency can mean better battery power,” Westneat said. “We’d really like to closely mimic the underlying skeletal structure and mechanical properties of the membranes and see if we can get super-high efficiency.”
Museums’ biological collections are another rich and underutilized resource for researchers. The Smithsonian, for example, holds nearly 600,000 specimens in its vertebrate collection alone, and Virginia Tech’s Rolf Müller has drawn upon these holdings for his work on bat-inspired drones. Using 3D scans of bat ears and noses from the Smithsonian, Mueller has created similar structures for his flying robot to help it report feedback through its zip-line-guided test runs.
“You have these millions of specimens lined up in drawers, which you can access very quickly,” Müller said. He’s been involved in the creation of a consortium of museum professionals and researchers to help make collections like these across the country more accessible for bioinspired advancement.
And then, no matter whether the source is swimming in a tank or lying in a storage drawer, translating that data into a useful form remains a challenge. “Your typical engineer wants specs, but the biologist might be handing them anatomical drawings,” Westneat said.
It wasn’t until he started going to some of these engineering talks himself that he realized his work could provide mechanical data of the fish’s movements that could translate into motor power and forces, data engineers need to produce a working machine. “Those are the things that natural selection can act upon, but they also make the difference between the autonomous vehicle that makes it back to the ship or not.”
Back to School
Learning, memory, and adaptation are other challenges entirely. Back at the Navy’s converted warehouse, the MeRLIn team is still primarily engaged with the problems of miniaturization. But they’re all too aware that the robot they envision wouldn’t be complete without the ability to learn, remember, and adapt.
Henshaw, who raises sheep at home when he’s not at the lab, said watching newborn lambs go from a moist heap to walking in a matter of hours underscores the difficulty of artificially replicating that process. “There’s no one who really understands how it works,” Henshaw said of the neural changes required of lambs to continually adapt their locomotion to rapid body mass changes as they grow into sheep. One approach his team is taking to address that strategy is to write software that allows them to change the way MeRLIn gaits are generated.
Separately, Henshaw is part of another project to develop a biologically inspired learning system. He showed me a video of a robotic leg kicking a ball into a small soccer goal. After three programmed kicks, the leg kicks the ball on its own 78 more times, systematically choosing its own targets and keeping track of its successes and failures. Further refined and applied to a robot like MeRLIn, code like this would make it easier for a walking robot to adapt on its own to different payload weights or leg lengths, for example.
“A lot of projects have equations that figure out how to optimize the center of gravity or motion through big mathematical equations in real time,” Henshaw said. “It works, but it’s not exactly biological. I can’t claim that the algorithm I’ve written [for the ball-kicking robot] is precisely what’s going on in the brain, but it seems like something that has to be going on. Humans learn to climb trees and kick balls through practice, not numerical optimization.”
Deep learning and access to collected knowledge would probably accelerate this process, Henshaw added, but there again, the hardware isn’t robust or small enough yet to fit onto something as diminutive as MeRLIn. “If you want these small robots, it’s not so much that we have to improve the algorithms but the hardware they run on,” he said. “Otherwise it’s going to take a computer that’s too big, with batteries that are too big, and it just won’t work.”
An Emerging Market
The shortcuts that biology provides for creating innovative body platforms and locomotion strategies may also help make biologically inspired robots more economically viable, as well. Choset is not the only academic who has started a company to help advance practical applications for his creations; in fact, Eelume, founded by Norwegian University of Science and Technology robotics professor Kristin Ytterstad Pettersen, is currently marketing its own robotic swimming snake for underwater exploration and inspection tasks. And De and Kinneally founded Ghost Robotics, a company to market Minitaur.
Large private companies are getting in on the game as well. Boston Engineering is in the end stages of running field demonstrations with its marine-inspection robot, dubbed BioSwimmer. This bot is not merely inspired by a tuna — its entire outer body is based on scans of a five-foot-long bluefin tuna that was caught near the company’s offices in Waltham, MA. And as with a living tuna, propulsion power originates in the tail, allowing the front half of the vehicle to be stacked with sensors and payloads. The goal wasn’t to mimic a tuna, though, but to harness the efficiency and high performance of the animal.
Mike Rufo, director of Boston Engineering’s advanced systems group, said the biological aspects of the design didn’t make it easier to build out, but it didn’t add extra difficulties either. Rufo claims the company built BioSwimmer (which is five feet long and 100 pounds) for about the same cost as similar projects — around $1 million — and that it will be priced similarly to other vehicles of its size. But efficiencies of motion provided by the tuna-inspired propulsion strategy allow it to operate longer on standard power sources.
“There are a few technical hurdles that are in our way, collectively, with bioinspired robotics,” Rufo said. “But bioinspiration offers opportunities to address those directly or to improve performance in a way that mitigates the impact of those challenges. For example, despite some really cool advances in battery technology, we’re on a plateau of how much power you can integrate into something of a given size. But if you can address the efficiency of a system, then maybe the battery doesn’t impact you so much. That’s one area where bioinspiration plays a big role.” Still, he thinks robots like these won’t be commonplace, in defense applications or otherwise, for at least the next five to 10 years.
Regardless of the monumental challenges that must be surmounted before we have not-too-creepy robotic helpers in our everyday lives, huge strides have been made even in the last several years toward encapsulating what biology and evolution have made clear: the dazzling ability of organisms to adapt and perform.
“It does seem Sisyphean sometimes, yes,” Westneat said. “I look at these aquatic robots, and they seem clunky to me; but then, I’m used to seeing these graceful animals swimming through a coral reef. But it’s not too outrageous to think that the engineers and biologists can get together and create robots that you throw into the water that swim off by themselves. Everything is exciting.”
Originally published at www.pcmag.com.