Earlier this month, a company named Agility Robotics unveiled its first ever robot: a bipedal creation named Cassie that looks like a headless, wingless ostrich. Cassie has reverse knees, motor-powered ankles, and can walk over different sorts of terrain at a decent clip. It can even survive a kick to the abdomen (the fastest way to test a robot’s self-balancing capabilities, though not the kindest). But are bipedal robots like Cassie really the future? Why not use wheels, tracks, or just more legs? Why make life difficult?

Well, according to Agility Robotics CEO Damion Shelton, there are good reasons for using bipedal bots, but it’s taken a while for the technology to catch up. He says the biggest advantage is that legged bots operate seamlessly in locations made for people.

“If you consider humans from a design standpoint, what we were designed for is being extremely agile in an extremely cluttered environment,” Shelton tells The Verge. He says when it comes to “legacy buildings” — i.e., those with stair-only access, or difficult steps or ledges — legged bots are going to be much more capable than those with wheels. “Or, if you want to be at ground level for the task you’re doing — like package delivery or on-site inspection.”

Shelton offers the example of 3D-scanning a rail yard. You could map a yard with a drone, but it would have to hover around, navigating in and out of buildings, and might require supervision. A wheeled or tracked robot would also have problems climbing stairs or making its way over uneven terrain. But a robot with legs would be as mobile as a human. Other use cases in a similar vein include scouting for the military and disaster response scenarios, like exploring a failed nuclear reactor or the epicenter of an earthquake.

“We’re not saying it’s the right solution for everything,” says Shelton. “In particular highly engineered environments, like inside modern factories where you need heavy lift capacity. Yeah, use wheels for that!” 

Though Agility Robotics is confident in the possibilities of bipedalism, the approach has unique challenges. When replicating human movement, engineers turn to the human musculoskeletal system as a model, swapping organic muscles for mechanical equivalents. At this point, says Shelton, we don’t have motors as strong or efficient as human muscles, and that makes the resulting robots slow, clunky, or reliant on external energy sources.

Take Boston Dynamics’ Atlas, for example. It’s not just bipedal, but humanoid, with arms as well as legs, all powered by hydraulics. Using hydraulics allows for precision and strength, but they can be difficult to control in way that fluidly mimics human movement.

Agility Robotics approach is to use an “under-actuated” design, meaning it has fewer motors than you’d need for full, like-for-like emulation. Instead, Cassie uses spring elements that mimic the passive feedback of organic systems. This lessens some of the challenges of controlling its behavior super-precisely.

“A good analogy is car suspension, where the mechanical design of the system determines a lot of behavior,” says Shelton. “Cassie will do a certain set of things mechanically without being told to by a computer. By having some springs in the system, we don’t have to handle impacts with the ground, for example.”

Sangbae Kim, who creates bio-inspired robots at MIT like this Cheetah bot, says that stability is still a big problem for bipedal bots. The likes of Cassie and Atlas might be able to survive the odd kick or poke, but ANY more than that and they’RE LIKELY topple over. Kim’s answer? Add more legs.

“Being quadrupedal is about having redundancies,” he tells The Verge. “Imagine you’re climbing and hiking in the mountain; in the most challenging situations you will use your hands. You won’t stay bipedal.” Kim says that humans probably became bipedal in order to free up their hands for using tools (it’s one of a number of explanations), and that four legs are just better suited for stability. “I strongly believe that quadrupeds have higher advantages when it comes to just mobility,” he says. “From leopards to mountain goats, quadrupeds can reach every corner of the land.”

Antoine Cully, a roboticist at Imperial College London, agrees that recent advances are making bipedal bots more viable, but says cost and robustness will continue to be a constraint in the near future. “They are very complex machines and this complexity increases their cost and their propensity to become damaged,” he says over email. “For these reasons, I think that the first commercial applications will be limited to domains that can support and afford these additional costs.”

Cassie is Agility Robotics’ first major prototype. The next iteration of Cassie will focus on bringing down the cost by increasing volume, with the firm aiming for a price tag “well under $100,000” in the next couple of years. The end goal is for a single Cassie to cost “less than a car.”

Currently, Cassie can reach speeds of around three meters a second and operate for 10 hours, shifting between walking and standing. The brains controlling the bot are dedicated to immediate movements rather than ranged navigation, but Shelton is confident those sorts of control functions can come later.

“I actually think we’re on the curve of something similar to self-driving cars,” says Shelton. “We’re probably not closer than four or five years to having bipedal robots go fully mainstream, but certainly not more than seven or eight.”