Should Robots Have Legs? An Investigation

Arguably, for most applications we might imagine, wheels are the way to go.

Michael Byrne

Michael Byrne

Image: Jordan Pearson/Motherboard

If you haven’t noticed, there aren’t a whole lot of wheels in nature. The best I can come up with are pillbugs, those weird little isopods that ball themselves up into protective orbs when threatened. The biological prohibition on wheels is pretty obvious. To spin independently, wheels need to be, well, independent of the whole. And that’s not how living things work.

Robots, however, have no such prohibition. Clearly, it’s far easier to give a robot wheels than it is to give a robot legs or wings. Locomotion via wheel is just a simpler problem in general. There’s no balancing involved (assuming four wheels or two sets of treads), and there are far fewer possible directions that force might be applied in order to move the robot. There are no joints.

But we still imagine—and build—robots with legs. Most robots have wheels, but that often seems more a matter of convenience than a particular design choice. The real-life legged robots that come most readily to mind are likely the Boston Dynamics family of Terminator dogs and monstrous would-be supersoldiers, or Agility Robotics, which is focused on developing walking robots that could one day deliver packages or help the elderly.

Robot legs are just a difficult engineering problem. Not as difficult as, say, biological wheels, but still. For one thing, slipperiness matters a great deal more to legged creatures than it does to wheeled machines. To that end, zoolologists are now proffering knowledge gleaned from cockroach locomotion to the robotics community, as in the case of a recent paper published in Frontiers in Zoology. They found that as cockroaches find themselves on slippery surfaces, their gait becomes desynchronized, switching from static stabilization to dynamic stabilization.

That really just underscores the difficulty of legged robots, so we might then ask, why bother at all?

The Army wants legs

Boston Dynamics robots largely come at the behest of military contracts. Big Dog, the quadruped robotics milestone-slash-meme, was built at the behest of DARPA with military ventures in mind, particularly those currently involving human soldiers. The Atlas humanoid robot, even more so.

The most general consideration when it comes to wheels vs. legs is terrain. Our whole awkward biomechanical setup as humans makes us very good at dealing with difficult terrain. We recreationally climb mountains and rock faces and have an entire sport that consists of running around cities and jumping off of staircases and such. We’re not as fast as some other animals and we don’t have crazy features like supersticky fingers, but we are pretty versatile.

The invention of the wheel wasn’t just the invention of the wheel, it was also the invention of the road. A key feature of most any large-scale military operation is roadbuilding. To wage war, troops and supplies need to be mobile, which requires wheels, which in turn require suitable surfaces to travel upon. As the US Army prepared for troop buildups in Vietnam in 1965, for example, road construction was prioritized only behind airfield construction.

What followed in Vietnam was pure screaming engineering hell. Sand was everywhere, and, when it comes to moving heavy equipment, sand is a highly unsuitable surface. Wheels sink in sand. To build the new military road network, Army engineers relied on rock. Lots and lots of rock that had to be mined in quarries, broken up by big machines, and then transported across, well, sand. "Rock was the word over there,” one commander said, according to a postwar engineering analysis. “I woke up in my sleep saying ‘Rock. rock, rock.' "

Legs can navigate through sand. They can navigate lots of things: steep hills, bombed out terrain, ledges. This ability to ascend vertical surfaces is crucial. Stepping over things, like mines or roots or battlefield debris, is also crucial. When it comes to getting up and-or over obstacles, wheeled robots have a fairly easy-to-see limitation: If the obstacle is more than twice as high as the robot’s front axle, the robot is pretty much boned. Treaded vehicles like tanks are better off, but not by much.

Boston Dynamics answer to this particular climbing problem is its lesser-known (and older) RHex robot—”devours rough terrain”—whose six half-circle legs allow it to very crudely approximate arm-assisted climbing.

But wheels are fast and cheap

You can really haul ass on a set of wheels. Motive power translates simply and efficiently from an engine to an axle. Keeping a four-wheeled robot upright, meanwhile, requires no complicated algorithms. Throw in some shock absorbers and dynamic suspension, and your wheeled robot can even navigate tough terrain, given the above restrictions.

Arguably, for most applications we might imagine, wheels are the way to go.

“The wheel has always been the easiest way to implement mobility in a vehicle, and also the fastest method of travel,” notes a paper from a trio of researchers at Sweden’s Mälardalen University. “Relative to speed it is also the most energy efficient way to travel. The implementation is often very simple, and does not require any advanced techniques such as vector controllers or additional joints to get the robot moving.”

Legs are a difficult algorithm problem. This difficulty generally has to do with a feature called “free gait,” which means that each leg of a legged vehicle or creature should be able to move independently. That means very fast, perfectly coordinated computations are required to keep the thing upright and moving in the right direction. Each one of Big Dog’s four legs includes four separate hydraulic actuators and two sensors. To be more useful than a wheel, a robotic leg requires at least three actuators/degrees of freedom, according to the Mälardalen paper.

And then there is the problem of how legs wield force in the first place. An improvement offered by Big Dog over prior legged robot attempts is its use of hydraulic actuators rather than independent electric motors for each joint. This makes it stronger, but also requires it to carry around a small go-kart motor (literally) to power the hydraulics. The Atlas robot, Big Dog’s bipedal kin, offers 28 actuators, but initially it had to be powered by a cord. The current Atlas iteration packs a lithium-ion battery pack. That’s good for just an hour of mobility.

Or maybe wheels attached to legs?

Sure, why not. That’s Boston Dynamics latest robot iteration, the aforementioned Handle. It’s a bit more utilitarian than its kin, or at least less sinister-seeming. It’s kind of like a robot on roller skates, with a pair of wheels mounted on simplified (compared to Atlas) legs. It goes about 9 miles per-hour and can jump up to four feet. I want to see what it can do in a skatepark.

Generally, Handle still needs roads. It’s not charging up and down wooded hillsides on battlefields. In the video above it’s clear that Handle is kind of approximating four wheels by using its arms to offer the precise counterweight needed to keep the thing upright.

The video also makes it clear that Handle does something different than offer a new sort of compromise between wheels and legs. This form of locomotion is more than the sum of its parts. This kind of motion feels fundamentally different, at least for a robot.

“This robot is using the momentum of its body. These dynamic movements are not what we know robots to be able to do,” robotics researcher Vikash Kumar told Recode last spring. “There are other robots that have the hardware capabilities of doing something similar, but on the algorithmic side, we’re not at the point where we can really leverage those capabilities.”

Nature doesn’t provide wheels, but that doesn’t mean it isn’t ingenious about mobility. Cockroaches are still just the beginning of what it can teach roboticists.