The Two-Legged Machine That Wants to Run Faster Than You: Inside Agility Robotics’ Roadrunner

A bipedal robot just ran a mile in under five minutes. Not on a treadmill. Not tethered to a safety rig. On open ground, outdoors, with nothing but its own two legs and an AI-driven sense of balance keeping it upright.

The machine is called Roadrunner, and it was built by Agility Robotics, the Oregon-based company best known for Digit, its warehouse-focused humanoid. But Roadrunner isn’t a warehouse worker. It’s a research platform — a stripped-down, 44-pound bipedal runner designed to push the boundaries of legged locomotion. And it just posted a time that would be respectable for a high school track athlete.

As reported by Digital Trends, Roadrunner completed a mile in 4 minutes and 58 seconds, reaching a peak speed of roughly 22 miles per hour during portions of the run. That’s not quite Usain Bolt territory — his top speed clocked in around 27.8 mph — but it’s faster than the average human can sprint and dramatically faster than any previous bipedal robot has managed over that distance.

The achievement matters for reasons that go well beyond bragging rights.

For years, the robotics industry has treated bipedal locomotion as a problem to be solved incrementally. Boston Dynamics’ Atlas could do backflips but was hydraulically powered, heavy, and operated in tightly controlled environments. Unitree’s humanoids have demonstrated impressive agility in short bursts. Tesla’s Optimus has walked, slowly, across stages at product events. None of them have done what Roadrunner just did: sustained high-speed running over a meaningful distance in an uncontrolled outdoor setting.

Agility Robotics designed Roadrunner specifically for speed. The robot stands about three and a half feet tall, has no arms, no torso to speak of, and no head. It looks less like a humanoid and more like a pair of mechanical ostrich legs bolted to a compact central frame. Every design decision — the carbon fiber components, the lightweight actuators, the minimal sensor payload — was made in service of one goal: move fast on two legs without falling over.

Why Running Is the Hard Part

Walking, from an engineering standpoint, is a controlled fall. One foot is always on the ground. Running is different. There’s a flight phase — a moment when both feet leave the surface and the robot is airborne. Managing that transition between ground contact and flight, thousands of times per mile, while maintaining stability on uneven terrain, is an enormously difficult control problem.

Roadrunner handles it through a combination of model-predictive control and reinforcement learning. The robot’s onboard computer continuously predicts its body state fractions of a second into the future and adjusts leg trajectories in real time. It trained extensively in simulation before ever touching pavement, running millions of virtual miles to develop the reflexes needed to handle perturbations — a gust of wind, a slight slope, an unexpected surface change.

The reinforcement learning component is particularly notable. Rather than programming explicit rules for every possible scenario, Agility’s engineers let the system discover its own running strategies through trial and error in simulation. The result is a gait that looks eerily natural. Smooth. Efficient. Almost animal-like in its rhythm.

Jonathan Hurst, Agility Robotics’ co-founder and chief robot officer, has spent more than two decades studying legged locomotion. His academic work at Oregon State University laid the theoretical groundwork for machines like Roadrunner. In previous interviews, Hurst has emphasized that the physics of running — the spring-mass dynamics, the energy storage and release in tendons — are universal principles that apply whether the legs are biological or mechanical. Roadrunner is, in many ways, a physical proof of that thesis.

So what’s the commercial angle?

Agility isn’t planning to sell Roadrunner units to Amazon warehouses. The robot is a research vehicle, a way to develop and validate locomotion technologies that will eventually flow into production platforms like Digit. But the implications are significant. A robot that can run at 22 mph can also walk with extraordinary stability. The control algorithms that keep Roadrunner balanced at speed will make Digit more capable on factory floors, construction sites, and disaster zones. The lightweight actuator designs could reduce costs and improve battery life across Agility’s product line.

There’s also a competitive dimension. The humanoid robotics space has become intensely crowded over the past 18 months. Figure AI raised $675 million and has been demonstrating its Figure 02 robot in BMW factory settings. Apptronik has partnerships with NASA and Mercedes-Benz. 1X Technologies, backed by OpenAI, is testing its NEO humanoid in real-world environments. Chinese firms like Unitree and UBTECH are iterating rapidly, with price points that could undercut Western competitors.

In this environment, a dramatic public demonstration of technical capability — a robot running a sub-five-minute mile — serves as both a recruiting tool and a signal to investors and potential customers. It says: our understanding of locomotion is deeper than anyone else’s.

The timing is deliberate. Agility recently opened its RoboFab facility in Salem, Oregon, which the company describes as the first factory purpose-built for manufacturing humanoid robots at scale. Digit units are already being tested by customers including Amazon and GXO Logistics. Demonstrating Roadrunner’s capabilities reinforces the narrative that Agility isn’t just assembling robots — it’s advancing the underlying science.

Not everyone is convinced that bipedal locomotion is the right approach for commercial robotics. Wheeled and tracked robots are simpler, cheaper, and more reliable in structured environments. Quadrupeds like Boston Dynamics’ Spot have proven effective for inspection tasks. The argument for bipedal robots has always been that human environments — stairs, doorways, cluttered floors — were designed for human bodies. A robot that moves like a person can operate in spaces built for people without requiring infrastructure modifications.

Roadrunner strengthens that argument. If bipedal robots can not only walk but run — quickly, reliably, over real distances — then the range of applications expands considerably. Search and rescue. Military logistics. Last-mile delivery in dense urban areas. Security patrols. Any task where speed and terrain adaptability matter more than raw payload capacity.

But significant challenges remain. Roadrunner’s 44-pound frame carries minimal payload. Its battery life at full sprint is limited. It has no manipulation capability. Scaling these locomotion achievements into a full-sized humanoid that can also carry objects, open doors, and interact with people will require solving a different set of engineering problems — thermal management, power density, sensor fusion, and dexterous manipulation among them.

And then there’s the regulatory question. A robot that can run 22 mph in public spaces raises obvious safety concerns. No regulatory framework currently exists for autonomous bipedal robots operating at speed alongside pedestrians. That’s a problem the industry will need to address before any of these machines leave controlled test environments.

Still, the Roadrunner milestone represents something tangible in a field that has been heavy on promises and promotional videos. A robot ran a mile. Outside. Fast. That’s not a rendering or a simulation result. It’s a measurable, verifiable achievement — the kind of concrete benchmark that the robotics industry has historically lacked.

For Agility Robotics, the message is clear: the hard problems in bipedal locomotion aren’t theoretical anymore. They’re engineering problems. And engineering problems, given enough talent and capital, get solved.

The race, it seems, is just getting started.