Ghost Robotics' Minitaur Quadruped Conquers Stairs, Doors, and Fences and Is Somehow Affordable

This robot uses innovative direct-drive motors to go pretty much anywhere

9 min read

Ghost Robotics Minitaur quadruped robot
Ghost Robotics' Minitaur quadruped robot.
Photo: Ghost Robotics

Bipedal and quadrupedal locomotion has been an ongoing challenge for robots. There’s been a lot of progress over the last few years, though, especially when it comes to dynamic motions: not just walking without falling over but also climbing, running, jumping, and more. This is the real value of legs: They enable robots to deal with the kinds of obstacles and terrain and situations that wheels and tracks can’t.

Getting quadrupeds to do these kinds of useful and fun things requires that a) you know what you’re doing and b) you have a robot that can do what you want it to do. Unfortunately, building legged quadrupeds is difficult, expensive, and time consuming. There is a small handful of bespoke research quadrupeds doing some very good work, but for the rest of us, having to actually do all of the hardware stuff is a major obstacle that makes it difficult to focus on the software, which is where the potential for real-world applications comes in.

Minitaur is a small but very capable platform that uses innovative direct-drive electric motors for a lot of power, virtual compliance, and integrated sensing.

In Professor Dan Koditschek’s lab at the University of Pennsylvania, Avik De and Gavin Kenneally put a lot of work into developing their own quadruped robot, called Minitaur. It’s a small but very capable platform that uses innovative direct-drive electric motors for a lot of power, virtual compliance, and integrated sensing. Minitaur was introduced this past July in an article in IEEE Robotics and Automation Letters, and there’s been enough interest in this little guy that Kenneally and De have started a company called Ghost Robotics to make sure that Minitaur is affordably available to anyone who wants one.

To understand why there’s enough demand for a small quadrupedal research platform like Minitaur to justify starting a company to build and sell them, just watch this video:

Well, that’s one way for a little robot to get through a human-size door. Wow.

What sets Minitaur apart from other legged robots are its direct-drive legs. Direct-drive means that there’s no gearing in between the electric motors and the legs and the ground, and there’s no elastic element either: It’s one rigidly connected system. This sounds bad, because compliance has been a big thing lately that everyone seems to want to put into their robots, and they usually do it with springs, or by relying on force or current sensors. Minitaur doesn’t use any of those, but the direct-drive motors can act compliant anyway, so you get the same benefits. You also get the advantage of being able to adjust every detail about the springy-ness in software, which is one of the reasons that Minitaur is such a useful research platform, as well as other advantages, like increased robustness, lower cost, and the ability to sense the ground through the motors.

For all of the details (all of them!), we spoke with Ghost Robotics cofounders De and Kenneally (who are also Ph.D. candidates at the University of Pennsylvania), and the company’s CEO, Jiren Parikh.

IEEE Spectrum: Can you tell me about Minitaur’s direct-drive legs?

Gavin Kenneally: In the 1980s, Harry Asada and Kamal Youcef-Toumi did a lot of research on direct-drive robot arms. For a robot arm, you can put heavy motors near the base so you don’t have to support their mass, and you can minimize the inertia of the arm. For a legged machine, it’s much more challenging because there’s no supported mass: You have to move your whole body weight whenever you walk around.

The problem with direct-drive legs is that you need to develop a lot of torque in actuators that need to be light and not heat up too much. By being very careful, limiting the number of degrees of freedom in the machine, and with an innovative leg design, we think we have the first direct-drive, legged robot. There’s also no gearbox, and that offers a number of advantages in terms of the robot being able to feel forces on the legs very quickly, and then also react very quickly. Also we’re able to do these really dynamic things without having to worry about destroying your gearboxes every time you have a hard landing.

Avik De:Minitaur first started based on a leg design Gavin was working on as part of his research at Penn. The design was meant to take advantage of some innovative kinematics; conventional leg design doesn’t allow nearly as much work space as the kind of atypical-looking legs that you see on Minitaur. It looks unconventional, and the reason is that it can access a much larger work space than more conventional-looking robot legs can. Putting four of those legs together with the custom-drive electronics, the direct-drive electric motors as described in the RAL paper, that’s basically Minitaur.

“This kind of spring-like behavior is really just the motors ‘impersonating’ springs . . . . We took the idea of a hardware spring and made it (using) software, so you get all the advantages of tuning. And, because the motors themselves are the springs, you know immediately once you’ve contacted something, and you get really good estimates about force and direction.”

With rigid legs, where does Minitaur’s dynamic motion come from?

Kenneally:The robot is very bouncy in its gait and looks very dynamic. It looks like the robot has physical springs, but actually the legs are made of rigid aluminum struts. This kind of spring-like behavior is really just the motors “impersonating” springs, and because they are direct drive and because the transparency is so good, it’s a very convincing impersonation: We took the idea of a hardware spring and made it [using] software, so you get all the advantages of tuning. And, because the motors themselves are the springs, you know immediately once you’ve contacted something, and you get really good estimates about force and direction.

So in addition to being actuators and springs, the leg motors are also sensors?

De:Yeah. Because the leg has very low inertia, the motors have very low inertia, and since the motors can get accelerated, you can detect that in software. So if you’re programming the robot, you can have a touchdown detection flag, which is exactly what we do with the bounding gait: Within 5 milliseconds of the physical event, or probably even less, you can get feedback of what kind of impact happens and what direction the force is, and you can use that in your controller.

What’s the bounding gait useful for, and what other gaits can Minitaur do?

De:The bounding gait is typically what animals use when they’re trying to travel at faster speeds because it enables the two hind legs to push together. It’s also fairly easy to model. If we were to approach rough terrain, which is something we plan to do in the future, we’d probably prefer to do something that doesn’t rely on reliable ground traction as much, something like a crawl. The other very interesting thing with the bounding gait is, if you attach a laser line scanner to the front of the robot, the vertical oscillations that are built into the bound gaits and the IMU that’s on the robot give you 3D point clouds for free.

Kenneally:Another point that Avik has touched on, in terms of augmenting gait, if you know that one leg is loaded more than the others, and we easily have that information, you can compensate for that. Or if the leg slips, you can compensate for that too. Or, if you’re trying to recirculate the leg, and you bump into something unexpected, you can sense and react very quickly. Turning up robustness to really rough terrain with these kinds of gaits is something we’re curious about.

The fact that you can just feel the contacts that the legs are making and react so quickly, with tunable compliance properties, really opens up a huge number of tasks that are possible with the robot; it’s just a matter now of figuring out the details and trying them. In terms of gait, you only see bounding in the video, but Avik also got it to do pronking, which is when the four legs are synchronized together, as well as trotting, which is when the diagonal legs are paired, so there’s huge versatility in terms of how the actuators and the legs can be controlled.

Ghost Robotics Minitaur quadruped robot jumpingMinitaur quadruped robot performing a 48-centimeter vertical jump.Image: Ghost Robotics

How does Minitaur compare to other dynamic legged robots, like MIT’s Cheetah?

De: We’ve worked with Sangbae [Kim] quite a bit, and his work has been very educational for us. [For Cheetah], he created these custom motors and custom drivetrains, and while Cheetah is able to do wonderful things, other people can’t really benefit from it yet.

Kenneally: MIT’s Cheetah has a single stage planetary gearbox. The robot is built to be high performance, but it’s also...I don’t want to say fragile, but there are a lot of parts that need to be carefully monitored to make sure it doesn’t break. On a smaller mass scale, with a 5-kilogram robot instead of a 30-kilogram robot, you inherently get a lot more robustness. We’ve been able to do these behaviors with a small machine that might be a little bit easier to do with a big machine, but with a small machine you’re really able to try things and if it doesn’t work, try it again.

How the heck did you manage to get Minitaur to open that door?

De:I don’t know if it’s clear from the video, but there’s a lot going on. The robot is jumping, it perceives that the door handle is there, retracts the leg, and manipulates the door handle.

Kenneally:Just to go over it in a little bit more detail: It jumps up on its front two legs, doing a handstand, and then jumps. The back left leg is waiting to feel the door handle, so it kind of sticks that leg out and waits until it senses contact. Again, all the sensing is through the motors, there’s no current sensors or force sensors. Once it perceives contact with the door knob, it retracts the leg, moves it over a little bit, and then extends it, and that actually all happens within 50 milliseconds, so it’s incredibly fast. And then once it’s done that, the other back leg, which is now also in the air, pushes against the door to crack it open a little bit, and it also helps push the robot so it pitches back down toward the ground, where it then retracts the leg back and catches itself before it falls. The door opening and stair/fence climbing were done with help from T. Turner Topping. We've just submitted a paper on this these behaviors, and Avik has a bound paper forthcoming as well.

Okay, so let’s say I’m sold. How do I get one of these?

Kenneally: We’ve been selling the robots for [US] $10,000. The one we’re currently making, it’s all very small-batch, so of course manufacturing is not scaled well.

Jiren Parikh:These are standard parts, off the shelf. If it can get up into a commercial environment with a prototype fairly quickly, manufacturing costs on this in volume could get to $1,500 a device. And probably less than that. That’s us just doing a rough estimate right now. What Ghost has done that’s unique is the fact that Minitaur is direct drive, and these machines can be ultralight and can be deployed at a very cost-effective, scalable model. 

“Legged robots have been starting to get more prominent. MIT’s Cheetah, Boston Dynamics’ Spot, and Spot Mini... But we’re able to bring some of the same capabilities in a way that’s interesting to researchers at a much lower price point, and people can access this much easier than if they were using completely custom designs.”

What kinds of potential commercial applications are you thinking about? What does the future look like here?

Parikh: Take a look at agriculture, for example. We’re seeing a huge amount of innovation in sensors, and there’s going to be more and more applications coming out to manage the actual plant genetics and so forth. Everything’s going to be based on either fixed sensors, or drones. But what we’re finding out, interestingly enough, is that there’s an opportunity for these low-cost legged robots out in the field as a 24/7 mobile sensor platform. We’re thinking about anything where you could use mobile sensors. Search and rescue. Military applications. Space exploration. There are scenarios where light weight, low cost, and very few moving parts could be fairly compelling.  

De: Legged robots have been starting to get more prominent. MIT’s Cheetah, Boston Dynamics’ Spot, and Spot Mini... But we’re able to bring some of the same capabilities in a way that’s interesting to researchers at a much lower price point, and people can access this much easier than if they were using completely custom designs.

Kenneally:It’s only a matter of time until legged robots are capable enough that they’re able to have a much stronger role in our lives. There are serious limitations to wheels and tracks in terms of both rough terrain and human-built environments. The complexity of legged machines is higher, but once we’re able to use them well, I think they’re absolutely going to be the go-to for robots that are able to help in the real world.

If this isn’t enough detail for you, their IEEE Robotics and Automation Letters paper, “Design Principles for a Family of Direct-Drive Legged Robots,” is available through open access over on IEEE Xplore.

Considering how relatively low cost Minitaur is, we’re expecting to see a whole herd of them wandering around out in the world soon, because if you have a robotics lab and you’re interested in legged robots, gait research, or just need a shiny new toy, 10 grand is one of those things that I bet you can just toss onto the end of some grant and label it “MISC. EXP.” or whatever and there you go. And based on what just one Minitaur has been able to do, it’ll be incredibly exciting to see what people come up with when these little guys are all over the place.

[ Ghost Robotics ]

The Conversation (0)