Humanoids and Legged Robots
Can a Humanoid Lift What a Human Lifts?
Published humanoid payloads of 16 to 25 kg look comparable to human limits, but the comparison collapses once posture, duration and repetition are included.
On paper yes, in practice no. Full-size humanoids publish 16 to 25 kg dual-arm payloads, which sits near the 23 kg reference load used in the widely applied NIOSH lifting equation. The comparison fails on posture and duration: the human figure is an ideal-conditions maximum that gets reduced by multipliers, while the robot figure assumes the load held close to the body and says nothing about holding it for a shift.
What a human can actually lift
The NIOSH lifting equation starts from a load constant of 23 kg and multiplies it down by six factors: horizontal distance from the body, vertical height at the start, vertical travel distance, asymmetry or twisting, lifting frequency and duration, and grip quality. Each is a fraction between 0 and 1, and they compound.
A worked case makes the point. Lifting from 25 cm in front of the ankles, from floor level to waist height, with a modest twist, four times a minute for two hours, with a fair grip, easily reduces the recommended limit to 7 to 11 kg. The 23 kg figure is a ceiling that almost no real task reaches.
What a humanoid can actually lift
| Condition | Effect on usable payload | Reason |
|---|---|---|
| Load held close to the torso | rated value | The datasheet condition |
| Load at arm's length | -40 to -70 % | Shoulder torque and balance moment |
| While walking | -20 to -50 % | Balance authority consumed by the payload |
| Reaching above shoulder height | -30 to -60 % | Unfavourable joint torque, higher centre of mass |
| Reaching below knee height | -20 to -40 % | Torso pitch shifts the centre of mass forward |
| Sustained hold | thermal limit | Static torque heats windings without motion |
The honest comparison
| Criterion | Humanoid | Human |
|---|---|---|
| Peak lift, ideal posture | 16 to 25 kg | up to 23 kg recommended |
| Sustained repetitive lift | limited by thermal and battery | 7 to 11 kg typical recommendation |
| Continuous working time | 2 to 5 h then charge | a full shift with breaks |
| Walking speed carrying a load | 0.5 to 1.2 m/s | 1.0 to 1.4 m/s |
| Awkward posture tolerance | Poor, torque limited | Good, but injurious |
| Consistency over 8 hours | Perfect until the battery ends | Declines with fatigue |
| Injury risk to itself | Mechanical wear only | Cumulative musculoskeletal |
The last two rows are where the actual argument lives. A humanoid does not get a back injury, and its 15th hour of a task is identical to its first apart from charging. Manual handling injuries are one of the largest categories of workplace lost time in most industrial economies, and removing the worst-posture tasks is worth more than matching a peak lift figure.
The task the robot should be given
Occupational data across industrial economies puts manual handling among the largest single causes of lost working time, and the injuries cluster on a small set of movements: lifting from below knee height, twisting under load, and repeated overhead placement. Those are precisely the movements where the NIOSH multipliers bite hardest, cutting a 23 kg allowance to single figures.
| Movement | Typical NIOSH multiplier | 23 kg becomes | Robot suitability |
|---|---|---|---|
| Waist height, close, no twist | 0.85 to 1.00 | 20 to 23 kg | Low value, humans do this well |
| Floor to waist, close | 0.55 to 0.75 | 13 to 17 kg | Moderate |
| Floor to waist with 45 degree twist | 0.40 to 0.60 | 9 to 14 kg | High |
| Above shoulder, arm extended | 0.30 to 0.50 | 7 to 12 kg | High, if the robot can reach |
| Repetitive, 4 per minute, 2 hours | 0.45 to 0.65 | 10 to 15 kg | Very high |
Read the right-hand column. The tasks worth automating are the ones where the human allowance drops to 7 kg to 14 kg, which is comfortably inside what a humanoid handles even after its own derating.
What follows for task selection
- Pick tasks by posture, not by mass. A 9 kg item lifted from floor level with a twist is a far better robot candidate than a 20 kg item lifted from a waist-height bench.
- Keep the load close. Design the task so the robot carries within 300 mm of its torso, which preserves most of the rated payload.
- Avoid sustained holds. Static torque produces heat without work. A task that requires holding a part steady for 40 seconds is harder on the machine than lifting it.
- Plan the charge into the takt. A shift is not a battery cycle, and pretending otherwise is how pilots miss their rate.
Frequently asked questions
How much can a humanoid robot lift?
Published dual-arm payloads run from 16 kg to 25 kg on full-size machines, measured with the load held close to the torso. At arm's length or while walking, usable payload falls by 20 % to 70 %.
Is that comparable to a human worker?
Only against the theoretical maximum. The NIOSH load constant of 23 kg applies under ideal conditions and is reduced by six multipliers, so real repetitive tasks often carry a recommended limit of 7 kg to 11 kg.
Why does walking reduce payload?
Because balance controllers use arm and torso momentum to generate correcting moments. A carried load adds mass and removes that control authority at once, so walking payload is always below standing payload.
Which tasks suit a humanoid best?
Awkward-posture, moderate-mass, repetitive tasks: lifting from floor level, twisting transfers, overhead placements. These are exactly the tasks that generate human injury, and mass alone is a poor selection criterion.
Can a humanoid work a full shift?
Not without charging. Published runtimes of two to five hours mean a shift requires opportunity charging, a pack swap or a second robot, and that has to be designed into the takt rather than discovered during the pilot.
Sources
- Applications manual for the revised NIOSH lifting equationNational Institute for Occupational Safety and Health, load constant and multipliers
- Unitree G1 entry in the ROBOTS guideIEEE Spectrum, published payload and runtime figures
- The Robot Report, humanoid hardware coverageReporting on published payload specifications and deployments