Meanwhile Google's Boston Dynamics test its new "Spot" quadruped -- should we pray for humanity... or welcome our new robotic overlords?

When Massachusetts Institute of Technology (MIT) Professor Sangbae Kim claims in journal papers to utilize "principles from biology" to build "bioinspired robots" [paper, abstract] it's far more than the usual funding-driven hype.  Kim's work has been worthy of dropped jaws in many cases.  And his latest effort in "bioinspired evolution" [paper, abstract] is downright frightening.  Kim's robo-cheetah is one cool cat and it's locked in a friendly feline feud with Google Inc. (GOOG) subsidiary Boston Dynamics.

I. The Road to MIT's "Biomimetic" Robo-Cheetah

You might recall how Google executive Andy Rubin departed from his role as Android OS commander-in-chief to build real life androids, of the karate mastering, terminator-like bipedal variety with a little help from MIT's Comp. Sci. and AI Lab (CSAIL).  Oh and remember Rubin's new lieutenant -- Professor Marc Raibert who headed up MIT's "Leg Lab".  Or more aptly, remember how his Google-acquired startup Boston Dynamics, Inc. (BDI) produced a steam-shooting robotic cheetah dubbed "Wildcat" that could outrun humans?

Well it turns out Professor Kim's Biomimetics Lab with MIT's MECHE (mechanical engineering) department has been hard at work fabricating a little leggy terror of his own.

Meet the Robo-Cheetah v.2.

MIT Robo-Cheetah v.2
MIT Robo-cheetah v.2

The design was born out of an amazing, if a bit fear-provoking, bit of friendly competition with the MIT-derived Boston Dynamics.  After three quarters of a decade in the books, both Prof. KIm's team at MIT and Google's Boston Dynamics have seen critical success, with each project earning a unique set of bragging rights.

The Boston Dynamics "Wildcat" and MIT "Robo-Cheetah" may look a bit like feline cousins, however there's subtle differences to note.  Perhaps most critically, while Boston Dynamics has taken an approach that leans more heavily on pure engineering, Prof. Kim has produced a much more organic looking robo-feline, thanks to a process he refers to as "biomimetics" -- mimicking nature.

The idea began some time back before he had hatched his plot to design diabolically fast cheetah-bots.  The year was 2008 and Kim was just finishing up his doctorate at Stanford University.  The media was buzzing that year over his "Geckobot", whose sticky fingers (and toes) helped it climb up vertical glass surfaces -- just like its biological counterpart.

Another DARPA-funded automaton, this robo-reptile showed off at that year's IEEE International Conference on Robotics and Automation 2008 (ICRA 2008), in Pasadena, Calif. where a paper Kim coauthored [paper, PDF] was a finalist for the best student paper award.  

Stanford Univ.
Before robo-cheetahs, Prof. Kim was making Geckobots (click to enlarge).
[Image: IEEE Transactions on Robotics/Stanford Univ./Prof. Sangbae Kim]

And another paper on the Geckbot which Kim was first author on garnered the King-Sun Fu Memorial Best Transactions on Robotics Paper Award that same year.  Kim received his PhD and spent the next year in a postdoctoral stint at Stanford's east coast friendly rival, Harvard University.

The key to the Geckobot was found in its success mimicking the superior designs observed in nature.  Kim and his colleagues used this process to tune the Geckbot's materials and mechnanics down to a microscopic scale.  Based on the success, Kim recognized the promise that a more thoughtful and intensive imitation of nature might hold.  He would soon put that vision to the test.

II. Something Feline This Way Comes

In 2009 Kim's flashy creations secured him the opportunity of an assistant professor assignment at MIT.  He named his new research lab the "Biomimetic Lab", opening a rich new chaper in the history of robotics at MIT that's already bred so many successes.  And it didn't take Kim long to set his sights on developing a robot that channels the cheetah, the world's fastest land animal.  

Cheetah, running
The cheetah is the world's fastest land animal. [Image Source: SantaBanta] 

The cheetah is capable of running short burst in speeds of up to 70 to 75 mph (112 to 120 km/h) and is capable of accelerating from 0 to 100 km/h (62 mph) in three seconds.  

Manmade quadruped robots remain far behind this type of speed.  The fastest robot to date has been Raibert's "Cheetah" design at Boston Dynamics, the treadmill predecessor to the free-roaming, untethered Wildcat.

Plans for the Cheetah were announced in 2011 and the prototype was complete by early 2012.  The bot built upon the success of Boston Dynamics' less speed cargo-hauling quadruped [PDF], BigDog which debuted in 2005.

The Boston Dynamic Cheetah is capable of running at speeds of up to 28.3 mph (45.5 km/h) -- a little less than half the speed of the real animal, but still faster than any human biped has been capable of running to date.

The Boston Dynamics "Cheetah" quadruped is seen in this 2012 shot setting the world speed record for a manmade bot.  [Image Source: DARPA]

The Boston Dynamics Cheetah was able to score the crown for the world's fastest robot thanks largely to its maker tuning the gait (running) algorithm to decrease inefficiency.  But a major part of Boston Dynamic's approach was to simply throw more problem at the problem.  The company describes:

[We] refined the control algorithms that coordinate the robot's leg and back motions and increased the installed power.

It's apparent that for all its power, one frustrating aspect of Boston Dynamics' "Cheetah" design is efficiency.  

And this isn't a new problem. Raibert's BigDog could jog at 4 mph (6.4 km/h) hitting a maximum range of roughly 12 miles after about 3 hours.  But it was also heavy, weighing 240 pounds (110 kg), and its 15-HP, 9000-RPM go-kart engine guzzled gas at a rate of 3 mpg to power its hydraulic legs.  BigDog had a published COT (cost of transport, a measure of efficiency of 15 J/(N*m).


Cheetah's COT (cost of transport, a measure of efficiency is unknown, but we do know that its power came from an external source.  WildCat reportedly used a similar 2-stroke go-kart engine to BigDog and weighed in at around 200 lb (90.7 kg).  Given the increased amount of force necessary to reach its top speed of 16 mph (25.7 km/h) at only a fifth lighter weight, it likely had an even more abysmal COT rating than BigDog.  That's a crucial reason why it can only gallop for 5 minutes before it runs out of the fuel stored in the small mounted gas tank.

Not all robots are this inefficient.  Honda Motor Comp.'s (TYO:7267) shuffling humanoid biped bot ASIMO achieved a modest COTS rating of 2.  Even more impressive the "Ranger" a shuffling quadruped bot developed by Cornell University in 2012 had a COT of 0.28 J/(N*m).  Why's that important?  "Ranger" was able to haul its 22 lb frame 40.5 miles in 30 hours and 49 minutes, setting a robotic distance record.  

Thanks to its more efficient electronic motors (which can reach up to 90 percent efficiency) it was able to consume on 16 watts of power during its marathon trek.  Its average speed was 1.3 mph (2.1 km/h).  The lesson learned from "Ranger" is that electrical motors trump gas-burning motors and hydraulics in efficiency, giving you a longer operational time and if you play your cards right, a better range.

Cornell Ranger
Cornell's Ranger holds the record for longest distance travelled by a robot.

But designing a Cheetah-speed robot with electric motors is a daunting task for several reasons.  First, electric motors are 90 percent efficient and tend to be high speed, but also are low torque severely constraining the design weight.  In order to get enough torque to drive even a modest weight, gearboxes must often dramatically reduce the speed to produce sufficient torque -- a key cause of Ranger's slowness.

In addition to slowing the speed, gearbox of electrical motors also reduce the efficiency.  On paper electrical motors use about 90 percent of the power put into them for work.  By contrast muscles typically are only about 20-25 percent efficient.  And yet due to the wastes required to matched legged land animals  in torque, legged robots have consistently been outperformed from an efficiency standpoint.  (See University of Michigan Mechanical Engineering doctoral student Brooke M. Haueisen's thesis [PDF] for a nice history of legged automatons.)

Humans, for example, can walk much faster than Ranger and have a COT of only 0.28 -- the best of both worlds.  Likewise the cheetah (the animal) can run ridiculously fast, but also enjoy a COT of around 0.5.

III. Something Fast This Way Comes

In an interview with IEEE Spectrum back in 2009, Prof. Kim first revealed his ambitions to make cheetah robots.  The paper wrote:

[Kim] is looking forward moving beyond cockroach-inspired design, and even past the gecko-inspired bots he worked on while at Stanford... [his] research at MIT is now focusing on "hyper dynamic locomotion." Translation? Cheetah-bot.

"You see so many robots everywhere," Kim says, "especially at conferences like these. But none of the robots can follow me, at walking speed, over rough terrain, or up stairs."

Kim thinks that kind of dynamic locomotion is "really limited still," and the robots that can crawl over rough terrain are small and often go slow. So he's aiming for a bigger, legged robot that could run fast, like a cheetah.

"Not necessarily over rough terrain," he says, "but over flat and also rough" spaces. And not necessarily even a cheetah, he says, not wanting to limit his goals. Squirrels, too, have highly dynamic behavior that allows them both to run fast and to climb, actions that can only be accomplished by two very different kinds of bots so far.

His goal crucially, was to design a bot that was not only extreme efficient like real life animals or Cornell's "Ranger", but also a bot with world record speed.  It was a tough challenge, but not an unfamiliar one for Kim's new employer.

The plan marked a return to roots of sorts for MIT, once home to Boston Dynamics founder Rabiert's famous Leg Lab.  Back in 1989 the MIT Leg Lab's "Planar Biped" robot (also known as the "3-D Hopper") set a speed record, sprinting at a brisk 21 km/h (13 mph).


The Planar Biped's record remains unbroken to this day for bipeds.  In terms of manmade robots of all varieties, only two quadrupeds (MIT's Robo-Cheetah and Boston Dynamics Cheetah) have faster recorded runs.

MIT Planar Biped
The MIT Planar Biped is enshrined in the MIT Museum. [Image Source:]

That record held up; two decades later when Kim started work on the MIT Robo-cheetah, the only robot managed to outspeed Raibert's biped -- the "Cheetah" quadruped he designed while at Boston Dynamics.

But after nearly two decades with little competition in the high speed department, Kim began to give his predecessor a real run for his money.  Funded by DARPA's Maximum Mobility and Manipulation (M3) program, Kim's Robo-Cheetah arose.  In spring 2012 he posted videos of the beast jogging at a respectable 5.4 km/h (3.3 mph).

A year later he cranked things up, sending the robotic feline galloping at a blistering 22 km/h (13.7 mph) on a treadmill.

MIT Robo Cheetah
Proud parent -- Prof. Kim poses with his robotic feline offspring back in 2013.
[Image Source: MIT Review]

That was fast enough to make it the second fastest robot in history -- behind only Raibert's "Cheetah".

And according to IEEE's Spectrum magazine, Kim realized not only his speed goals for the robotic feline, but his efficiency goals as well.  To realize high torque, Kim and his colleagues eschewed Boston Dynamic's bulky gas-powered hydraulics, instead opting for a lean design based on electric motors.  To solve the torque riddle they built their own specialized coaxial high torque electric motors, in effect mimicking naturally occuring muscles.  The approach reduced waste, but more work was still needed.

Robo Cheetah motors

Also onboard were circuitry to regenerate power via the kickback that occured when the robot's foot pushed off the ground.  Similar in spirit to regenerative braking, this could eventually be added to automobile struts to improve battery life in mild and battery electric vehices.

The Robo-cheetah uses kevlar tendons to dampen impact stress and its coaxial motors are tied to circuits that regenerate part of the return force experience upon the foot hitting the ground.
[Image Source: Sangbae Kim/MIT/Sage]

To further improve the efficiency of his runner, Professor Kim outfitted it with nature-inspired designs, including kevlar "tendons" and faux spine w/ polyurethane rubber discs.    He also made sure the bot had a bounding gait, as small, forceful jumps have beens shown to be mechanically superior to shuffling gates (hence why running animals bound or gallop).  These additions, along with a carefully crafted algorithm would eventually deliver a COT of 0.52.

Robo Cheetah

Robo cheetah

That was a stunning achievement and one that was underappreciated by all but a few.  While the number got cursory media attention back in 2012, most just looked at the speed numbers, saw that it was slower than Boston Dynamics' Cheetah and let out a yawn.

But consider first that Kim's Robo-cheetah v.1 achieved a level of efficiency nearly comparable to a real life cheetah and only half as inefficient as the super slow "Ranger" robot.  Now consider that versus BigDog-- and almost certainly Wildcat, Kim's Robo-cheetah might have had a two order of magnitude advantage in its COT rating.

And in addition to being somewhere on the range of 50-100 times more efficient, it was nearly as fast as Wildcat, falling just barely short of Cheetah's on-treadmill record.

With that great efficiency, comes flexibility and perhaps greater commercial potential (or for those worried about robots going terminator, greater risk).  The Robo-cheetah even in its first incarnation could spend hours traveling at modest speeds under its own power -- something Wildcat was utterly incapable of.
MIT Robo Cheetah
[Image Source: MIT Review]

In a 2013 ICRA conference paper [PDF] the authors report that the first generation "cheetah robot" was capable of trecking at a rate of 8.3 km/h (5.2 mph) for 1.23 hours.  Alternatively, quanitfying the endurance by range, it could walk for 10 km (6.2 miles).  And it did all of that on only a 3 kg (6.6 lb.) battery pack.

This was a robot that might finally be ready for the real world.  Prof. Kim explained in a 2013 press release:

In order to send a robot to find people or perform emergency tasks, like in the Fukushima disaster, you want it to be autonomous.  If it could run for more than two hours and search a large field, that would be useful. But one of the reasons why people think it’s impossible to make an electric robot that does this is because efficiencies have been pretty bad.

In a corresponding university press release  he further added:

With our system, we can make our robotic leg behave like a spring or damper without having physical springs, dampers or force sensors.  The majority of impact energy goes back to the battery because the damping is created by custom-designed electric control of the motor.  [The motor] regenerates energy that would have been lost.

It was an impressive and underappreciated start, but work still remained.

IV. Improving on Greatness

As 2013 drew to a close, Professor Kim and his team turned their focus on crafting a second-generation successor to the materful initial prototype.

To create the Robo-Cheetah v.2, Professor Kim and his team would reach out to a number of collaborators to perfect the design.  MIT Electric Engineering Professor Jeffrey Lang worked closely with David Otten, a principal research engineer in MIT’s Research Laboratory of Electronics (RLE) to perfect the high-torque motor design and maximize regenerative energetic kickbacks.

Professor Kim meanwhile perfected the gait, determining that by exerting a small amount of extra force when the leg struck, a smooth motion could be achieved.  The speed was found to be directly correlated to the force exerted on the ground as the foot struck.  

Robo cheetah
Robo Cheetah gait

Gait images set
Developing an optimal gait and understanding the source of energy losses was key focus of Professor Kim's work. [Image Source: Sangbae Kim/MIT/Sage]

Professor Kim notes that animals (and humans) who run fast don't let their feet spend much time on the ground (they are said to have "short" duty cycles), but exerts a lot of force while it's there.  He remarks:

Many sprinters, like Usain Bolt, don’t cycle their legs really fast.  They actually increase their stride length by pushing downward harder and increasing their ground force, so they can fly more while keeping the same frequency.  Bounding is like an entry-level high-speed gait, and galloping is the ultimate gait.  Once you get bounding, you can easily split the two legs and get galloping.

Most robots are sluggish and heavy, and thus they cannot control force in high-speed situations.  That’s what makes the MIT cheetah so special: You can actually control the force profile for a very short period of time, followed by a hefty impact with the ground, which makes it more stable, agile, and dynamic.

As a result, they’re way louder.  Our robot can be silent and as efficient as animals. The only things you hear are the feet hitting the ground. This is kind of a new paradigm where we’re controlling force in a highly dynamic situation. Any legged robot should be able to do this in the future.

Oh great, so it can sneak up on us, terrific news.  The finished design was unveiled by the team in Sept. 2014.

Robo cheetah

Robo Cheetah

The new model was capable of walking comfortably or galloping outdoors on grassy terrain thanks to its gait control algorithms, which resemble those of Boston Dynamics WildCat.  On the treadmill it proved itself capable of comfortably stepping over "slight bumps in its path" such as "foam obstacles".  It even was able to jump.

MIT Robo Cheetah

Robo cheetah v2
MIT's Robo-cheetah v.2 [Image Source: AP]

Professor Kim's students and colleagues, including research scientist Hae-Won Park and graduate student Meng Yee Chuah, presented a paper on the new bounding algorithm was published in the Int'l Journal of Robotics Research, and a presentation on it was given at the IEEE/RSJ International Conference on Intelligent Robots and Systems in Chicago, Ill.

V. That's One Small Step for a Cheetah, One Giant Leap for Robo-Cheetah Kind

Buzz built around the bot yet again last week after MIT release a video and article detailing the robot jumping over barriers 18 inches (0.46 m) high -- approximately half its body height.

The team had been working to perfect this new trick since late 2014...

The jumping algorithm uses a new onboard sensor -- a LIDAR (Light Detection and Ranging) sensor produced by Japanese parts company Hokuyo Automatic Comp., Ltd.  Using that laser sensor, the Robo-cheetah develops a map of the plane ahead, representing the ground as line.  Objects that stick up from the ground in 3-D space are identified by their height above the ground and how far from the bot they are.

Once the bot has selected an object to jump over, it calculates an approximate point to best leap from and further optimizes that jump point via an "approach adjustment algorithm".  Once at the intended spot, the robot performance a leap by more forcefully kicking up off the ground. The force and angle are generated via the jumping algorithms third and final stage, a trajectory calculation.

Robo Cheetah hurdle

Robo cheetah leap

Robo cheetah hurdle

While the Biomimetic team has typically been laser-focused (no pun intended) on efficiency, when designing the jump code, it wisely put utility first.  Comments Prof. Kim in a press release:

A running jump is a truly dynamic behavior.  You have to manage balance and energy, and be able to handle impact after landing. Our robot is specifically designed for those highly dynamic behaviors.

If you want to optimize for, say, energy efficiency, you would want the robot to barely clear the obstacle — but that’s dangerous, and finding a truly optimal solution would take a lot of computing time.  In running, we don’t want to spend a lot of time to find a better solution. We just want one that’s feasible.  We’re too obsessed with optimal solutions. This is one example where you just have to be good enough, because you’re running, and have to make a decision very quickly.

The algorithm was codeveloped by MIT research scientist Hae won Park and Patrick Wensing, a postdoc in Prof. Kim's lab.  It will be put to the ultimate test later this week at the 2015 DARPA Robotics Challenge, which runs June 5-6 in Pomona, Calif.  The event is expected to see a leap-off between Boston Dynamics' latest quadruped and Robo-cheetah v.2.

Robo cheetah
Park, codeveloper of the jumping algorithm is pictured working on Xbox controls for the Robo-cheetah.
[Image Source: AP]

Following the DARPA event, the trio will in July detail the algorithm and project progress at the 2015 Robotics: Science and Systems conference at Sapieza University in Rome, Italy.

What's next after that?  It remains to be seen.  According to Prof. Kim, another speed run may be in order.  The Robo-cheetah v.2 is expected to top out at 30 mph (48.3 km/h).  If it can hit or beat that mark it may be able to seize the world record from Google's Boston Dynamics.  Suffice it to say that's more than fast enough to keep up (or should we say overtake?) any human alive.

MIT Robo Cheetah v2

Robo Cheetah
[Image Source: AP]

Looking further ahead, Prof. Kim has stated he wants his Cheetah-bot project to ultimately hit in 35-40 mph ( km/h) range (with a Robo-cheetah v.3 third-generation design, perhaps?).  If that kind of speed can be realized while maintain the MIT Robo-cheetah v.2's superb efficiency, it would truly be a daunting challenge to Boston Dynamic's WildCat and kin.

VI. Boston Dynamics Spot Eyes Friendly Feline Feud with MIT's Cool Cat

Speaking of the devil... er robo-feline, Wildcat (release in Oct. 2013) has evolved.  In February of this year, Boston Dynamics paraded out "Spot", a downsized evolution of Wildcat is seen in a video walking alongside BigDog's packmule successor, LS3.

Much like MIT's Robo-cheetah v.2, Spot uses LIDAR to "blindly" navigate its enviroment untethered.  However, one key difference is while the MIT Robo-cheetah v.2 is still human operated (with algorithms filling in the blanks in turning simple steering commands into a stable gate and leaps, when desired)... 

MIT Robo Cheetah
The MIT Robo-cheetah v.2's controls are mostly on the operator side. [Image Source: MIT]

...Spot appears to be semi-autonomous with its control electronics housed in an onboard brain (uh oh).  Using its smarts, Spot can climb stairs, creep through forests, resist mankind's best attempts to kick it to the ground thanks to its advanced electrical controls and hydraulic actuator technology.

Boston Dynamics

Efficiency-wise Spot improves things in part by cutting the weight to 160 lb (72.6 kg).  And more importantly, RoboHub reports that the petroleum burning engine and hydraulic mechanics have been swapped out for a more efficient equivalent, making Spot a bit closer to the MIT Robo-cheetahs in internals.

Boston Dynamics

Almost certainly this means that Boston Dynamics' bot will see a quantum leap in efficiency over WildCat.  The real question is whether that fresh-found efficiency will come at the cost of speed and if so how much.  In its press video, Boston Dynamics clearly showed that Spot is nimble and is sophisticated in its terrain navigation and stability algorithms.  It also appears to have a longer battery life, likely a half hour or more.

However, the fastest the robot was shown trotting at the pace of a human jogger, likely around 6 mph (9.7 km/h)  Will Spot be able to keep up with the sleek robo feline produced by Professor Kim's lab.  Boston Dynamics' world speed record may be depending on that.

Spots vs. MIT Robo Cheetah

But perhaps the more compelling question is which quadruped wins the contest for most terrifying limbed robot of 2015?

Is it Spot, the robot that can creep through the forest to disrupt your pleasant picnic?  Or is it Prof. Kim's Robo-Cheetah v.2 who could probably use that jolly jumping to pounce on you as you fumble over fences with your clumsy human appendages?

Let's face it, both of these robots -- funded by money from the U.S. military's research wing -- would be pretty terrify if one day their future successors turned against us.  Mankind will have to keep a wary eye on them, even as they're deployed for new operations, including disaster response, hauling gear on the battlefield, and perhaps even acting as combatants on battlefields of the future.

Sources: MIT [press release], YouTube

"It looks like the iPhone 4 might be their Vista, and I'm okay with that." -- Microsoft COO Kevin Turner

Latest Blog Posts

Copyright 2017 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki