Digit’s run shows what humanoids must prove before the warehouse shift

Agility Robotics’ video, "Innovation at Agility: Running", shows Digit facing one of the simplest and least forgiving tests in humanoid robotics: the robot has to run, not just walk. The company’s own description keeps the claim grounded. Digit has not yet had to run during customer deployments. So this is not a new warehouse feature being marketed as operational necessity. It is a test of the boundary between simulation and a physical machine.

That distinction matters. Digit is positioned as a work robot for spaces built around people, not as a lab prop for one impressive motion clip. In the source text, Agility repeats the core product argument: it makes robots "made for work," and Digit works alongside people in human-designed spaces. Its role is to handle tedious, repetitive tasks suited to a machine, leaving companies and workers focused on work that requires judgment, coordination and responsibility.

Running therefore is not just a viral extra for the capability list. In a humanoid robot, every increase in speed exposes physics that a polished walking demo can hide: foot contact, mass transfer, balance correction, actuator response time and tolerance for small errors in pose estimation. If the locomotion policy was trained in simulation, the real question is not whether it looks convincing in a virtual model. The question is whether it survives when the physical robot has friction, inertia, latency and small imperfections in its hardware.

Agility’s description points directly at that problem: the only way to prove whether simulated policies translate is to do it. That is the sim-to-real issue without unnecessary theater. Simulation is used in robotics because it is faster, cheaper and safer than repeatedly crashing expensive hardware. But simulation is not the world. If a control policy only behaves well inside a model, its value is limited. If it can drive a physical Digit into a run, then at least part of that model has started to pay rent in reality.

The limits are just as important. The video does not prove that Digit will be sprinting through distribution centers tomorrow, or that customers are asking for that. Agility Robotics publicly frames its machines around work deployments, and that focus is clear on the company’s official Agility Robotics site. Running, in this context, is a robustness test rather than a direct operating plan. In warehouse and logistics settings, stability, safety and predictability often matter more than speed.

Still, the demonstration has technical weight. Humanoid robots become serious when they stop being a sequence of isolated poses and start behaving like systems that can change movement regimes without losing control. Walking, accelerating and running are not merely three positions on the same speed slider. They are different balance regimes. If Digit can move into more dynamic motion, Agility is showing that its development program is measuring more than whether the robot can complete a task. It is probing how much physical margin the system has before that task becomes risky.

That makes the short video more useful than a typical robotics spectacle. It does not tell us that the humanoid worker has suddenly become a sprinter. It tells us that Agility is testing the question that will determine how useful such robots can be outside perfectly controlled conditions: whether behavior learned in simulation can remain stable once metal, sensors and the floor take over.