Watch enough humanoid robot demonstration videos and a pattern emerges that nobody mentions in the press release: at some point, the unit stops, walks to a charging station, or gets rolled off camera. Every demo ends. The robot doesn't go home, make dinner, and come back tomorrow. It gets plugged in. And the question of how long it can run — and how much energy it consumes doing so — turns out to be one of the more underexamined constraints on when and where humanoid robots can actually be useful.

Battery life rarely appears in the headline metrics companies choose to publicise. Speed, payload capacity, degrees of freedom (the number of independent joints that allow a limb to move), dexterity benchmarks — these make good video. A robot's runtime per charge, or its energy consumption per task cycle, does not. But for any operator trying to build a business case around deploying humanoid systems, the power question is unavoidable.

The Numbers That Are Actually Available

Published battery specifications for commercial humanoid robots are sparse, and the ones that exist require some unpacking.

Agility Robotics' Digit has a published runtime of roughly two hours per charge under working conditions. Boston Dynamics hasn't released official Atlas battery specs for its electric-platform version, but earlier hydraulic Atlas units required a tethered power supply for extended operation — a constraint that heavily shaped where and how they could be demonstrated. Unitree's G1, which costs around $16,000 and targets the research market, specifies approximately two hours of runtime. Figure AI and 1X Technologies have not published detailed battery specifications for their commercial platforms.

Two hours is the number that keeps appearing in this space, and it deserves some context. A human warehouse worker operates for an eight-hour shift, with breaks, reliably. A robot running two hours per charge either needs a very fast charging cycle, multiple units rotating through a single task, or a docking system that lets it recharge during natural workflow pauses. None of those solutions are impossible — but each adds complexity and cost that rarely factors into high-level comparisons between robot and human labour.

There's also the question of what "two hours" means under load. Like a smartphone battery rating, runtime under laboratory conditions and runtime during heavy physical work are different figures. A robot carrying a tote, navigating a real warehouse floor, and constantly reorienting its cameras and sensors will drain power faster than one standing in place during a demonstration. Published specs tend not to specify conditions clearly, which makes direct comparison between platforms difficult.

Why Bipedal Robots Use So Much Energy

The power demands of humanoid robots are partly a fundamental physics problem, and understanding it helps explain why the constraint is harder to solve than it might appear.

Walking on two legs is energetically expensive. Humans manage it efficiently because evolution had hundreds of millions of years to optimise our locomotion — our gait recycles mechanical energy through tendons and muscles in ways that reduce the net metabolic cost. We're also good at standing still with minimal effort, because our skeleton locks into place without continuous muscular activation.

Robots don't have tendons in the biological sense. Most current humanoid platforms use electric actuators — motors driving joints — that require continuous power draw to hold a position, not just to move. Every joint that's not perfectly balanced is drawing current to stay upright. The more joints (modern humanoids have 30 to 50 or more), the more standby power draw. Add locomotion, manipulation, onboard computing for vision and motor control, and the communication systems needed for remote oversight, and the power budget adds up quickly.

Some platforms use hydraulic actuation instead of electric motors. Hydraulics can deliver high force with relatively compact actuators, but hydraulic systems require a pump running constantly and typically need an external power supply for extended operation — which is why early Boston Dynamics Atlas units were tethered in many demonstrations. The shift to all-electric designs across most commercial platforms is partly driven by the desire for untethered operation, but electric actuators bring their own power management challenges.

Passive dynamic design — building mechanical systems that exploit gravity and momentum to reduce powered movement — is one engineering approach to improving efficiency. Some research platforms have demonstrated genuinely efficient bipedal walking using this principle. But passive dynamic designs tend to be fragile and task-specific, which conflicts with the general-purpose flexibility that makes humanoid robots commercially attractive in the first place. The efficiency gains come at the cost of versatility.

The Charging Infrastructure Question

A robot with a two-hour runtime in an eight-hour shift needs a workable answer to the charging question, and that answer has implications that extend beyond the robot itself.

The straightforward solution is hot-swappable batteries — units with modular battery packs that can be swapped in under a minute, the way some power tools work. Several humanoid platforms are moving in this direction. It works, but it requires a supply of charged backup batteries, a maintenance worker or automated system to perform the swap, and downtime during the swap itself. For a single robot, this is manageable. For a fleet of dozens of units, battery logistics becomes a genuine operational consideration.

Autonomous charging docks — where the robot navigates to a charging station during natural workflow pauses, similar to how a robotic vacuum returns to its base — are another approach. This works well for tasks with predictable downtime windows. It works less well for tasks where continuous operation is the value proposition.

A third approach, longer-term, is simply better batteries. Energy density in lithium-ion cells has improved by roughly 5 to 8 percent annually over the past decade, a pace that is meaningful but not dramatic. Solid-state batteries, which promise significantly higher energy density without the flammability risks of liquid electrolyte cells, have been widely anticipated for several years. They are making real progress in automotive applications, and the humanoid robotics industry will benefit when they reach commercial scale — but the timeline is not certain, and the improvement may not be as large as optimistic projections suggest.

What This Means for Deployment

The practical consequence of current battery limitations is that humanoid robots are better suited to some deployment contexts than others, and the constraint shapes which industries will see early adoption.

Tasks with natural pause cycles — where a robot can dock for fifteen or twenty minutes during a shift without disrupting workflow — are more compatible with current runtimes than tasks requiring continuous eight-hour operation. Tasks in environments where charging infrastructure can be built in from the start (a new warehouse designed around robotic operation) are easier than retrofitting charging into an existing facility. Tasks that run in a single, bounded area are simpler than tasks requiring a robot to range across a large site.

This partly explains why tote handling in a dedicated section of a fulfilment centre is a more tractable first deployment than, say, general-purpose construction work. It's not just task complexity — it's that warehouse environments can be designed to accommodate the robot's operational constraints, including power, in ways that unstructured outdoor environments cannot.

Healthcare and elder care, which often appear in discussions of humanoid applications, present a particularly difficult power challenge. Care tasks are physically variable, often time-sensitive, and take place in environments where running dedicated charging infrastructure is inconvenient. A robot that needs to charge for two hours after two hours of work is a different proposition in a hospital corridor than in a controlled manufacturing bay.

The Comparison That Doesn't Get Made

There's a framing that shapes a lot of humanoid coverage that the battery problem quietly undermines: the idea that a single robot can replace a single human worker on a one-for-one basis.

A human worker arrives, works eight hours, goes home. A robot with current battery technology either needs multiple units to cover an eight-hour slot, a logistics operation to keep it charged, or it works less than a full shift. None of those scenarios are disqualifying — but they do change the unit economics calculation in ways that tend not to appear in the projections companies publicise.

The honest framing is that battery limitations are a real constraint on deployment economics today, and will remain a real constraint for at least the next several years. The companies making genuine progress on this — through better battery chemistry, smarter energy management software, or operational designs that work with current runtimes rather than against them — are solving a problem that the demos don't show. That makes it easy to overlook. It doesn't make it unimportant.