Pneumatic Actuators Give Robot Cheetah

Electric motors have helped bring legged robots into the mainstream, offering a straightforward and compact way of controlling robotic limbs with all the fancy control features you need for safe and nimble motion. What you can’t get out of electric motors (more than once, anyway) is the kind of instantaneous power needed to match the performance of biological muscles. This is why Atlas, arguably the most powerful and dynamic robot out there right now, uses hydraulic actuators—to make a human-size robot do a backflip, that’s just about the only way of getting the power you need.

Inspired by the high-speed maneuvering of cheetahs, roboticists at the University of Cape Town, in South Africa, have started experimenting with the old-school sibling of hydraulic actuators—pneumatics. By using gas as a working fluid instead of a liquid, you can get a high force-to-weight ratio in a relatively simple and inexpensive form factor with built-in compliance that hydraulics lack. Are pneumatics easy to control? Nope! But to make a robot run like a cheetah, it turns out that complicated control may not even be necessary.

“We’re arguing that fine force control is maybe not needed for rapid maneuverability.”—Amir Patel, University of Cape Town, South Africa

First, let’s talk about what’s wrong with hydraulics—because hydraulics are complicated, expensive, and all kinds of messy if they ever explode, which they sometimes will. And while the noncompliant nature of hydraulics makes them easier to model and control, it also makes them less forgiving in real-world use. If you go back far enough, to the 1980s, when Marc Raibert was developing dynamic legged robots at MIT, those running and jumping robots were relying on pneumatics rather than hydraulics, because pneumatics were much easier to implement.

One big reason why everyone seems to be using hydraulics rather than pneumatics nowadays is that air is compressible, which is great for built-in compliance but messes up most traditional control methods. “Fine force control is difficult with this actuator, and most have avoided it,” explains Amir Patel, an associate professor at the University of Cape Town. “Hydraulics is not compressible and can do amazing things, but it’s quite a bit more expensive than pneumatics. And when looking at animals that require explosive motion from their limbs, we thought that pneumatics would be a good, and often overlooked, actuator.”

Patel has done an enormous amount of research on cheetah biomechanics. We’ve written about some of it in the past. (For instance, here’s why cheetahs have fluffy tails.) But recently, Patel has been trying to find ways to track cheetah dynamics in very high fidelity to figure out how they’re able to move the way they do. This would be easy if the cheetahs would cooperate, but from the sound of things, trying to get them to run directly over a small force plate or do the maneuver you want while in ideal view of the cameras you’ve set up is kind of a nightmare. Much of this work is ongoing, but Patel has already learned enough to suggest a new approach to cheetah-inspired locomotion. “From our years studying cheetahs here in South Africa, it appears as if they’re not really trying to do fine force control when accelerating from rest,” Patel says. “They’re just pushing off as hard as they can—which makes us think that an on/off actuator [also known as a bang-bang controller] like pneumatics could do that job. We’re arguing that fine force control is maybe not needed for rapid maneuverability tasks.”

“We focus on the transient phase of the locomotion—like rapid acceleration from a standstill, or coming to rest once you’re at a high-speed gait.”—Amir Patel, University of Cape Town, South Africa

Patel (along with colleagues Christopher Mailer, Stacey Shield, and Reuben Govender) has built a legged robot (or half of a legged robot, anyway) called Kemba to explore the kind of rapid acceleration and maneuverability that pneumatics can offer. Kemba’s hips incorporate high-torque quasi–direct drive electric motors at the hips for higher fidelity positioning, with high-force pneumatic pistons attached to the knees. While the electric motors give the kind of precise control that we’ve come to expect from electric motors, the pistons are controlled by simple (and cheap) binary valves that can either be on or off. The researchers did put a lot of effort into modeling the complex dynamics of pneumatic actuators, because you do after all need some understanding of what the pneumatics are doing. But again, the concept here is to use the pneumatics for explosive actuation and get finer control from the electric motors at the hips.

Kemba, the two-legged (and boom-stabilized) robot, uses electric motors for precision and pneumatics for fast movements.University of Cape Town, South Africa

With a boom for support, the 7-kilogram Kemba is able to repeatedly jump to 0.5 meters with a controlled landing, and it reaches a maximum jump height of 1 meter. While it’s tempting to focus on metrics like jump height and top speed here, that’s really not what the research is necessarily about, explains Patel. “With Kemba (and all the robots and animals we study in my lab) we focus on the transient phase of the locomotion—like rapid acceleration from a standstill, or coming to rest once you’re at a high-speed gait. Most papers don’t really concentrate on that phase of the motion. I would love for more labs to be publishing their results in this area so that we can have some metrics (and data) to compare to.”

Patel would eventually like Kemba to become a platform that biologists could use to understand the biomechanics of animal locomotion, but it’s likely to remain tethered for the foreseeable future, says first author Chris Mailer. “A lot of people have asked when we will build the other half or if it is realistic for Kemba to carry around a compressor. While this would be awesome, that was never the intention for Kemba. The main objective was to execute and learn from bioinspired motions rather than focus on onboard power or autonomy.”

This doesn’t mean that Kemba won’t be getting some upgrades. A spine could be in the works, along with a tail, both of which would provide additional degrees of freedom and enable more dynamic behaviors. There’s a long way to go before legged robots get anywhere close to what a real cheetah can do, but the pneumatic approach certainly seems to have some promise. And anything that has the potential to lower the cost of legged robots is fine by me, because I’m still waiting for one of my own.