MIT’s Robotic Cheetah Is Getting Even Scarier

Robotic Cheetah get's upgraded bounding algorithm

Researchers over at MIT are hard at work upgrading their Robotic Cheetah. They are developing an algorithm for bounding movement, after researching how real cheetahs run in the wild.

Mach 2 is fully electric and battery-powered, can currently run at speeds of 10MPH (however they’re predicting it will be able to reach 30MPH in the future), and can even jump over obstacles 33cm tall.

We originally saw the first robotic Cheetah from Boston Dynamics in cooperation with DARPA two years ago — it could run faster than any human alive (28.3MPH) but in its tests it was tethered to its hydraulic power pack and running on a treadmill. It’s unclear if MIT’s Cheetah is a direct descendant from that one, but they are both supported by DARPA.

The technology in this project is nothing short of amazing — its electric motors are actually a custom part designed by one of the professors of Electrical Engineering at MIT, [Jeffrey Lang]. In order for the robot to run smoothly, its bounding algorithm is sending commands to each leg to exert a very precise amount of force during each footstep, just to ensure it maintains the set speed.

Did we mention it can jump over things too?

34 thoughts on “MIT’s Robotic Cheetah Is Getting Even Scarier

  1. But can it steer? Keeping the leg pairs in sync it may as well be an inline two leg robot. The motion it’s missing is the bending of the spine, which cheetas use to get more force pushing with their hind legs and to be able to take longer strides and bounds to run faster.

    1. Can please not point the things out that it currently is not able to? This list would be long and even in 50 years there will be things we can point out what a robot can not do. We can all assume that the developers are aware that their robot currently can not steer. But they have to start somewhere and walk this road over years. There are currently 4-legged robots out there that can already steer; But can not jump. Everyone is focusing on different things.

      I bet that steering in on their list. Also: higher jumping, agile movement, better vision… but again: give them a break, they have to start somewhere. This is already way more impressive then the last 5, 10 and 20 years.

    2. I’m with you on the spine. It caught my eye that the robot has no spine so it seems that the engineers had to compensate making the entire gait look awkward. Checked a couple of slowmo vids of Cheetahs running and they have an unbalanced gait. They “favor” a side as humans do so the left or right side will hit the ground first, even in full stride. Makes sense, from a full stop, one would push off with both pairs at the same time to get the mass moving. Once in motion then it’s more efficient to offset the power in each pair to keep the mass moving because, you know, Newtons first law and all. Looks like a spine is kind of important for that gait.

      Still… it is very impressive and is a forgivable offense for a prototype. I would like to see them work on the spine in the next revision. The tail is a requirement in that formula too.

      1. They could probably use a spring, like the buggy suspension the Corvette uses, for a spine. With 2kW of power to store in it they should have more than enough energy at hand for some huge leaps, literally and figuratively.

    3. If you had read the actual articles, you’d know that they’ve only just developed the bounding motion of the cheetah that it uses to pick up speed before transitioning to a gallop with alternating legs.

    4. Looks like it would barely be able to steer and if it could, would probably have to do so at a near full stop. Tho I could be wrong. While it is impressive the fixed leg position (insert animal) running on a treadmill seem to have been done to death. It’s almost analogous to the Hello World of (bi/quadra)pedal robotics. Tho I do give them some serious props for actually integrating its power supply and showing it operating without the safety tether at the end of the clip.

      Either way it isn’t a small feat. I know I will be really impressed when someone turns out a robot that can corner as well as it can run in a strait line and jump over hurdles. Not to say one doesn’t exist just that I haven’t seen it.

      1. Yes, plus in most movies people never think to run left or right – only straight ahead.
        I think the feeble human resistance will be crushed as long as they stick with this thinking.

  2. If this is meant to be suggestive of a cheetah, then it must have been modelled on a rather handicapped cheetah as those hind legs are really hobbling. Look at the point at which it jumps – it appears most of the jump effort is done by the font legs, the rears barely stretch out.
    And as another poster has mentioned, all the footage is of it moving in straight lines. I can only suspect they must not have the cornering problems resolved or the video would have shown it turning about, moving in circles and so on.

    1. It has no eyes and no tail to keep balance. It simply can’t turn.

      The main point of the robot is to develop torque/force control of the contact point with the ground instead of static balance and seesaw action by attempting to keep the geometric center of mass pivoting around the contact patch like the Big Dog robots do.

      In other words, this robot is leagues closer to actual dynamic walking and running than any other walking robot out there because it doesn’t try to know and predict the exact geometry of the robot relative to the ground, but rather infer what it should be doing by feeling what sort of forces act on its legs and joints like biological beings do because we haven’t got precise position sensors in our joints or laser rangefinders for eyes.

    2. The Big Dog robot for example always plants two diagonal legs on the ground to form a seesaw based on where it guesses its center of gravity is and where inertia is pulling it. Then it begins to tip over around this pivot, and if the guess was correct it will fall diagonally forwards. If the fall is unexpectedly slow or in the wrong direction, it revises its guess about the center of mass for the next step.

      Once it senses that it is falling in the right direction at the right speed, it catches up with the other diagonal leg pair and lifts the first two, and falls forward again but diagonally in the other direction. Do it fast enough and it goes forward in a slight zig-zag motion. That’s why it’s always trotting even when not going anywhere. If it puts all four legs on the ground, it no longer “knows” where its center of mass is and can’t easily get moving again.

      It’s unlike what any animal actually does. It computes exactly where it needs to place each leg every step to form another diagonal pivot to keep falling in the right direction.

      In contrast, we plant our legs roughly where they should go, and then apply a horizontal force on the contact patch to propel us forwards. We also use the pendulum action to save energy, but without computing every step. It becomes very apparent if you unknowingly step onto a patch of ice. You slip because you’re applying too much lateral forces for what grip you have.

    1. thats an example of static stability.

      making a dynamically stabilized robot turn probibly just requires the right algorithms. applying slightly more force on one side at the start of a stride should induce a yawing motion on the robot, which over the course of several strides would turn the robot. doing this while maintaining stability on the other exes may require roll compensation gyroscopes to cancel out any unwanted roll (pretty much preforming the same job as a tail).

  3. Thus far this is outwardly an electric version of the Boston Dynamics projects – if the researchers are going to claim similarity to cheetah movement (which they have), they’d have to at least get to the point of leg overlap – other comments note some of the flexibility and tail issues that give the cheetah its extraordinary burst speeds. The real progress seems to have been a more arcane redesign of the stability and propulsion systems that are much less apparent and not explained well.

    More philosophically, this is an unfortunate example of the expanding role of claims bloat in research – a very good project done by brilliant people that has to prove its relevance by extrapolating to less-certain future uses in order to prove relevance (and likely to maintain funding).
    http://images.nationalgeographic.com/wpf/media-live/photos/000/004/cache/cheetah-run_494_990x742.jpg

    1. “a very good project done by brilliant people that has to prove its relevance by extrapolating to less-certain future uses in order to prove relevance”

      You might notice that the researchers themselves talk about the electric motors and the amplifiers used to control them, while everyone else who are doing the second hand reporting are putting the emphasis on the fact that it’s a cheetah robot.

      Do not confuse research publications with popularization of science and mis-interpretation by the media. It’s downright depressing how many people think reading an article on New Scientist or something means they’re reading the actual science and what the actual research is saying.

      1. I think we’re talking around the same point – overhyping of research – but in the first video the researchers themselves, particularly Dr. Kim, talk about using robots for transportation without roads and exceeding existing muscle capabilities among other things as a result of their research.

        1. But that’s what DARPA is actually developing them for. They want to use quadruped robots as supply mules and logistics vehicles where it would be too difficult to drive a truck.

          They also want to make them stronger than biological animals, because otherwise it would just make more sense to use actual mules.

  4. I must say i like the “actual cheetah” tag in the vid i almost confused them before it poped up. …. any way this is a great step forward in robotics i can see a smaller version of this being in stores next Christmas as a kids/hackers toy.

    On a side note a cheetah stretches its spine as it bends back this spring tension is then released rapidly to the feet. this device runes like no natural animal I know closest would be a spring bock or elk.

    1. Perhaps instead of immediately going for cheetahs they should be a bit more realistic and go for the movement of for instance a rhino, they are heavy and relatively stiff but can run and are still slightly more flexible than this robot.
      And let’s face it with current available powersources you can expect things to be heavy.

  5. The thing is, you can design your own electric motors all day long but there are simply limits to an electromagnetic system, you can only do so many windings in such space, and even if you supercool it (bit hard in an independent robot) you still run into limitations of physics I expect.

    But since they improved on the available products I guess it shows we didn’t reach the limit of design, but still though, I’m not sure you can upscale it much beyond this, and obviously they want faster and higher jumping and bigger and able to carry stuff.

    1. Addendum: I say windings but obviously there are also limits to the reaction and falloff times of strong electromagnetic fields that at obey certain laws, but that is actually interesting to research since there might well be more ways to counter such effects somewhat in ways we never tried, or collectively forgot about – which is something that also seems to happen on a regular basis.

Leave a Reply

Please be kind and respectful to help make the comments section excellent. (Comment Policy)

This site uses Akismet to reduce spam. Learn how your comment data is processed.