0018-9162/04/$20.00 © 2004 IEEE48 Computer

C O V E R F E A T U R E

P u b l i s h e d b y t h e I E E E C o m p u t e r S o c i e t y

Biomimetic Robots

F rom the ominous Klaatu of The Day theEarth Stood Still to the Terminator, we’veseen robots typically portrayed on screen asstiff, humanoid machines. But it’s not justHollywood that has locked robots to the

human form.“A lot of conventional thinking pervades the field

of robotics,” says Morley Stone, a program man-ager in the US Defense Advanced Research ProjectsAgency’s defense sciences office (www.darpa.mil/dso/). “They still look very much like they aredepicted in grainy black-and-white films. You seethis humanoid robot that doesn’t walk very well.We still haven’t improved upon that all that much.”

Forget the anthropomorphs. Today, researchersare looking in the cupboards of their local dinersand under rocks for biological inspiration to createa new generation of flying, crawling, and swimmingautomatons known as biomimetic robots. Intriguedby how other species have adapted to a whole worldof environmental niches, researchers are workingto understand and reverse-engineer the adaptivetraits of creatures, including those—like the seem-ingly indestructible cockroach—we might muchrather step on than study.

MIMICKING BIOLOGYBiomimetics is a general description for engi-

neering a process or system that mimics biology.The term emerged from biochemistry and appliesto an infinite range of chemical and mechanical phe-nomena, from cellular processes to whole-organ-ism functions.

“People have been trying to copy nature for a verylong time,” says Jerry Pratt, a research scientist atthe Institute for Human and Machine Cognition(www.ihmc.us). Leonardo da Vinci made drawings

of potential flight contraptions based on detailedanatomical studies of birds, and the Wright broth-ers based their airplane structure on observationsand analysis of bird flight. However, researchersdiverge in precisely how they define biomimetics.“‘Biomimetic’ is often a vague term, much like‘robot,’” says Pratt.

Mark Cutkosky, a professor in Stanford Uni-versity’s Department of Mechanical Engineering, ispart of a team working on a family of legged robotsbased on cockroach locomotion. He says their teamdefines biomimetics as “extracting principles frombiology and applying them to man-made devices—particularly robots.”

Cutkosky says two forces are driving the “newwave” of robotics. First, biological research hasexposed a huge amount of biological process datathat roboticists can apply to their work. Second,advances in low-cost, power-efficient computingsystems allow researchers to create robots thatwork outside laboratories. Cutkosky says thatroboticists can “really put some of the lessons we’relearning from biology to practice. Ten years ago,even if I had understood exactly what materials andmechanical principles underlie the cockroach’srobust dynamic locomotion, I would have beenunable to build a robot that embodied them.”

Not that current biomimetic robots are de-pendent on the fastest computing technologies available.

“The interesting thing about the biomimeticwork,” says Butler Hine, manager of the comput-ing information and communications technologyprogram based at NASA-Ames, “is it uses nature’sevolved way of doing things rather than the com-putationally intensive way.” In lieu of algorithmic-intense artificial intelligence, Hine says, some

A wealth of biological data together with advances in low-cost, power-efficient computer systems support the emerging development of robotsthat mimic insect and sea creature adaptations to environmental niches.

Linda DaileyPaulson

September 2004 49

researchers are using control loops and 8-bitprocessors and field-programmable gate arrays(FPGAs) for command control rather than lines andlines of programming.

Biomimetic robots are still relatively new, how-ever, and the possible collaborations among biolo-gists, robotic engineers, and computer scientistshave barely begun.

There’s more to this process than simply con-structing a workable, autonomous robotic device,say scientists. “How birds fly, how fish swim, howdolphins locate objects, and how humans walkmight best be discovered and understood by tryingto reproduce these activities in a device,”contendsIMHC’s Pratt. “The knowledge gained might notbe immediately useful, but it could some day leadto useful technologies based on, but not necessar-ily mimicking, these phenomena.”

RESEARCH PROJECTSMost of the current robotic projects sprang from

DARPA programs, says Hine. The US is the pri-mary financial underwriter for research throughDARPA and other agencies such as NASA, theOffice of Naval Research (ONR), and the NationalScience Foundation (NSF).

Researchers have hopes of creating robots thatcan detect mines, explore Mars, or search for peo-ple trapped beneath an earthquake-damaged build-ing. It is premature to predict which of the manyexisting projects will be widely deployed first; thevariety of concepts and potential applications bothalone and in combination with other roboticsresearch is simply still too broad.

Sprawl hexapodsCutkosky is part of a team at Stanford’s Center

for Design Research (www-cdr.stanford.edu/biomimetics) that is designing and fabricating six-legged robots that “draw their inspiration from thephysical construction and mechanical design prin-ciples that are responsible for the robustness of thecockroach.” Funded primarily by ONR, the groupincludes Bob Full, a respected biology researcherfrom the University of California, Berkeley, whosework on the mechanics of cockroach locomotionunderpins the robot design.

Several characteristics of cockroaches intriguethe Stanford team, including the speed and stabil-ity with which they can negotiate rough terrain.“They run over obstacles without slowing down orgetting knocked off course, and they do this mainlyby virtue of having a wonderful tuned mechanicalsystem—sort of like the suspension of a car—that

keeps them stable and on course,” says Cutkosky.Plus, he notes, “It’s hard to damage a cockroach.”

Figure 1 shows a robot from the Sprawl family.What makes these robots different, says Cutkosky,is their mechanical, rather than computational, prop-erties. “In the past, legged robots were expensive andrequired fast computation and accurate sensors toachieve rapid locomotion. In contrast, Sprawl robotsrely on a tuned, resonant mechanical system.”

The system’s six legs move in an alternating-tri-pod gait in a “sprawled” design that mimics thecockroach’s biological structure and both supportsthe robot and allows it to move fast. The system isoperated using an open-loop motor pattern drivenby a clock associated with the on-board processor.The various sensors, actuators, and microproces-sors are embedded in the robot’s durable polymershell, made possible by a complex fabricationprocess called shape deposition manufacturing.

Cutkosky says that government funding is essen-tial because the applications are still a few yearsaway. More work is required to make the robotsmore robust and to improve fabrication. He wouldlike to make a version using injection-molded plas-tic parts and subject it to testing in real-world appli-cations, such as military reconnaissance.

Adds Cutkosky, “We are not trying to ‘copy acockroach.’ This would be impractical. Andbesides, who would want one?”

Robotic lobstersFor years now, two independent teams of

researchers—one concentrating on sensory-chem-ical tracking and the other on locomotion—havebeen working toward creating a chemical-track-ing, underwater robot based on lobster biology.

Neuroscientist Frank Grasso, an associate pro-fessor of psychology at Brooklyn College, has been

Figure 1. iSprawl robot. The small (115 mm long, 315 grams) cockroach-inspired robotruns autonomously at 15 body lengths per second (photo courtesy of Stanford Centerfor Design Research).

50 Computer

part of a research team examining the lobster’sacute sense of smell in the turbulent ocean envi-ronment. Grasso headed the robotics group thatdesigned and built two generations of autonomousunderwater robots, Robolobsters I and II, whichmatch the size as well as the sensing and locomo-tor capabilities of their biological counterparts.These robots let researchers test hypotheses basedon controlled observations of a real lobster’s super-human ability to detect olfactory information andmake decisions based on it.

Because Robolobster operates under water, anytether to power interferes with operations, requir-ing researchers to make it as close to autonomousas possible. Grasso says that Robolobster II is com-putationally and energetically autonomous for mis-sions up to five hours.

Another team of researchers is working on mak-ing the first robot with an elementary nervous sys-tem. This project stems from research on lobsterand crayfish nervous systems conducted in the1970s by Joseph Ayers, a biology professor atNortheastern University (www.neurotechnology.neu.edu/). Ayers subsequently used biosonar teleme-try to study lobster behavior in the wild.

When DARPA approached him in the 1990sabout building a robot, Ayers quips that he was “acard-carrying neurophysiologist.” He assembled agroup of biologists, naval architects, and electricaland mechanical engineers, several of whom, he says,were “experts on what’s impossible.” Figure 2 showsa prototype robot currently under development.

In an important evolutionary piece of roboticresearch, Ayers is collaborating with the Universityof California, San Diego’s Institute for NonlinearScience to create a locomotion control system thatdoes not use typical motors and finite-statemachines as controllers. The team is working to

reduce the electronic neurons and synapses to ana-log VLSI and to generate “motor program-like cen-tral pattern generators based on a nonlineardynamic model of real lobster neurons,” Ayersexplains. “The artificial neurons generate actionpotentials that gate power transistors to drive arti-ficial muscle.” This eliminates the need for a feed-back loop in a motor controller, as many robotcontroller systems require. Modulation of chaos inthese networks will enable more animal-like behav-iors in robots, such as a squirming motion.

Eventually, Ayers wants to create an artificialbrain by integrating gravity, bump, and flow sen-sors with the central pattern generators the re-searchers have already developed, thereby formingan elementary nervous system.

Robotic lobsters have been funded by an alpha-bet soup of agencies, including DARPA, ONR, andNSF. “The only delay is to really find a mission [forthe technology],” says Joel L. Davis, an ONR pro-gram officer working on adaptive neural systems.Davis expects one lobster to be available for use bysummer 2006.

One military application is minesweeping beach-heads, but the robots must have specific knowledgeabout their mission, says Davis. They must know,for example, if they should detect mines that areon the ocean floor or buried beneath its surfaceand, when they are working in a swarm, whetherto communicate with each other or not. If one lob-ster detects a mine, it could, for example, signalothers in the area to go away before it detonatesthe mine, taking itself out at the same time.

EntomopterAt Georgia Tech, work continues on Ento-

mopter. This tiny robot is designed to both crawland fly, but its name stresses its flying ability.

Aerial robots have existed for about two decades,says Robert Michelson, a Georgia Tech professorand principal research engineer, but “they don’tnecessarily take on biological form.” As Figure 3shows, these robots look more like other machinesthan winged animals because “the bioinspiredthings are not as well understood.”

When the Entomopter research was initiated inthe mid-1990s, the idea was to design a micro-sizedair vehicle about the size of a military MRE (meal,ready to eat) and sturdy enough to survive a GIaccidentally sitting on it. The dream device couldgo over a hill or obstacle and “find bad guys ..., butit’s unrealistic,” says Michelson. Factors such asdelicacy and the need for line-of-sight communi-cation made it impractical.

Figure 2. Robotic lobster. This robot prototype uses biomimetic control principles. Its behavior is based on a library of action patterns reverse-engineered from lobsterbehavior in the target operational environment (photo courtesy of ONR).

September 2004 51

Subsequent research proved “size doesn’t mat-ter” for outdoor operations, so Michelson says hisgroup shifted focus to niche indoor operations. Histeam is working on devices nimble enough to entera building through a chimney or open window, flyfast, evade detection on camera surveillance, andnegotiate tight areas. Such a device would be usedfor reconnaissance and for missions such as dis-rupting electrical equipment. And, perhaps, theymight even be mistaken for a large moth.

This generation of the Entomopter is designedfor operation in two atmospheres: a 50-gram ter-restrial version and an aerospace version designedfor use in different gravitational environments.Both versions are constructed primarily from car-bon composite material. The design feature thatintrigues aeronautical engineers is a circulation con-trol process that turns high-speed, hot-gas flow intoa lower speed, cooler gas that can, when vented outthe wing, cause flow that gives the vehicle seventimes more lift and lets it fly at slow speeds. Perfectfor exploring Mars.

Michelson expects computer scientists to even-tually help create a fully autonomous device, butother problems are more pressing. Once the scien-tists resolve the flight mechanics, they can work onflight control. Until they know whether the device“turns on a dime or on a quarter … it doesn’t makesense to do flight algorithms.” Besides, processorsand other computing devices will change manytimes before the Entomopter is deployed,Michelson says.

DARPA has provided much of the funding, andNASA is interested—but it will probably be six ormore years before Entomopter is deployed on aMars mission.

Bugs and WhegsBiomimetic research at Case Western Reserve

University began in 1987 with insect behavior stud-ies that employed neural networks on biologicaldata. Today, the Biologically Inspired Robotics Lab (http://biorobots.case.edu) creates machinesinspired by nature.

Roger D. Quinn, professor of mechanical engi-neering and the lab’s director, says inspiration is amore accurate description of their work since mim-icry is neither possible nor desirable. He says robotsshould not be restricted by an animal model’sdesign as is the case with airplanes, which can flyfaster and carry payloads heavier than the birdsthat inspired their development.

Quinn’s team has been working on robotsinspired by cockroaches and crickets as well as a

hybrid mechanism called Whegs (wheels plus legs).Whegs is the device that is closest to commercialdeployment, says Quinn. This fairly simple robotuses one motor and “needs little software in termsof locomotion.”

Figure 4 shows a cricket-inspired robot, approx-imately three inches long, designed for both walk-ing and jumping. Quinn says that one of the mostpromising projects for this device involved soundtracking in collaboration with Barbara Webb, anexpert in artificial intelligence and biology at theUniversity of Edinburgh.

Figure 3. Entomopter. The multimodal design is adapted for indoor flight operations. Its wings beat autonomically from a chemical energy source (photo courtesy of Georgia Tech).

Figure 4. Cricket-inspired robot on a 2-inch grid. The robot can both walk and jump tonavigate terrain with features much larger than itself (photo courtesy of BioroboticsLab/Case Western Reserve).

52 Computer

Crickets use sound to track potential mates. Theteam is investigating this phenomenon for robotsthat could be used in search and rescue efforts. Theidea is to deploy small robots in lieu of dog orhuman search units to safely locate cries for helpor detect breathing in a constricted environmentsuch as ruined buildings following an earthquakeor explosion. The design places sensing micro-phones close together like cricket ears so the robotcan use the Doppler effect to locate sound.

Quinn’s lab has also applied this idea to a full-sized Whegs robot. He says combining the two sys-tems “just makes sense.” Applying the idea tomini-Whegs is also a possibility. This three-inchversion of the platform can run up to 10 bodylengths per second and comes in a jumping versionas well.

Like many other programs, Case Western robotshave been funded by DARPA and other US mili-tary research agencies. Quinn says work on theWhegs platform continues because “the military isgoing to fund something they can use as soon aspossible.” A next step is to eliminate radio controlin favor of autonomous operation.

Germany’s Scorpion The German-led Scorpion project (www.ais.

fraunhofer.de/BAR/SCORPION/) is creating arobot for use in environments where humans eithercannot or do not want to go, says BernhardKlaassen, a researcher on the team from FraunhoferInstitute. The team, led by Bremen academic FrankKirchner, chose the scorpion to emulate because itis “fast, robust, and in some sense kinematicallycomplex,” Klaassen says. One of the more notablephysical attributes people remember about a scor-pion is the tail, with its venomous stinger. Versionsof the robot use the stinger to transport a tiny cam-era rather than a painful payload.

Figure 5 shows one of these eight-legged robots.Successive Scorpion generations have employed anincreasing number of sensors and more sensor datato help them move smoothly. “To read and inter-pret all these sensor inputs, we used not only a fastprocessor on board but also a programmable hard-ware device, an FPGA, to get all sensor inputs pre-pared within the 100-Hz control loop,” Klaassensays. This enables the Scorpion to, for example,increase current to the motor to push away a stoneor use a higher swing motion to help its leg clearan obstacle.

“Our latest developments are more concernedwith neural control for walking robots,” he says.Small, recurrent neural networks and artificial evo-lution help the robot “learn” simple rules such ashow to navigate, including, for example, how to getout of a corner. “The interesting feature of these net-works is robustness,” says Klaassen. “If you trans-fer the identical net to a completely different robotthat only knows how to change its direction to leftor right and how to ‘see’ a wall, it will react in a sim-ilar way if the situation is similar. But you never haveto explain what a corner is and what to do then.”

Klaassen says Scorpion, funded primarily byDARPA, is still a research platform, but its learn-ing capability is an important part of the autonomythat many military projects require.

RESEARCH DIRECTIONS AND APPLICATIONSAlthough the funding has vanished for some

promising projects, including Case Western’scricket, Morley Stone expects DARPA’s investmentin biorobotics to continue, as it is “so important inmany aspects of what we do across the Departmentof Defense.” He especially sees the need for missionsinvolving reconnaissance and defusing explosives.

New ideas are still eliciting funds from US mili-tary research agencies. For example, an octopus-inspired project is looking at creating soft arms withsuckers that can bend in any direction. Grasso’sgroup at Brooklyn College is part of an interna-tional team that also includes researchers fromHebrew University, Penn State, and Clemson.Grasso is enthusiastic sbout the work, saying “it’sgoing to keep us going for a few years.”

Corporate funding, apart from Japanese compa-nies such as Sony and Fujitsu, has been negligible,say researchers. The lack of commercial activity ispartly because the field is very young, but NASA’sHine expects pieces of biomimetic research to begradually introduced into mass-produced commer-cial devices by much the same process that resultedin fuzzy logic being used in vacuum cleaners.

Figure 5. Scorpion robot. The 60-cm, 9.5-kg robot integrates a robust network for navigational and other rules of learning (photo courtesy of Fraunhofer AIS).

It seems that some of this research is destined foruse in toys that will appear on the commercial market, not unlike Sony’s robot dog Aibo. Toyapplications are actually very challenging, saysStanford’s Cutkosky. “The companies that aremaking toy robots have to do extremely clever engi-neering to achieve entertaining performance at anacceptable cost. I think if you asked iRobot engi-neers about the challenges associated with Roombaversus their expensive military robots, you’d findthere were many,” he says.

University research has resulted in some spin-offcompanies founded by former students. Stanfordspawned Iguana Robotics, for example, which ismaking a cat-inspired robot. Hine says some stu-dents who have completed advanced studies in theUS have returned to their home countries andopened businesses. Hungary’s AnaLogic Com-puters Ltd., founded by a Berkeley graduate, is oneexample.

Japan is probably the first nation in which robotassistants will be accepted, but other nations mayslowly accept this technology into the mainstream.“If robots are more clever and more helpful in pri-vate houses, then of course, the companies willjump on it,” says Klaassen. “It would be a sad

thing,” he adds, “if we had only military robotsand not friendly ones.”

C omputer science is a critical tool for both biol-ogists and roboticists in this enterprise. “Ifyou’re a biologist, you start to simulate these

things in hardware or software,” to gain betterunderstanding, says ONR’s Davis. And if you’re aroboticist, “Once you get a beating wing, you haveto learn how to control it.”

The bulk of the research work ahead is concen-trated on making robots autonomous, and as thiswork continues, researchers expect collaborationwith computer scientists to increase.

The interdisciplinary nature of the work isintense. Because each creature is so exquisitelymade and so vastly different, it can be difficult forteams of biologists, chemists, and engineers tounderstand it, much less devise a facsimile. Whilethe progress of biomimetic robots from the labo-ratory to the unpredictable world we live in mayseem slow, from the perspective of evolution, it maybe on a pace to beat a cockroach. �

Linda Dailey Paulson is a regular contributor toComputer. Contact her at ldpaulson@yahoo.com.

September 2004 53

Visit Computermagazine online for current articles,links to onlineresources, and a collection of classics that changed the computing field.

www.computer.org/computer/