Robot ROI and Cost per Hour: Total Cost of Ownership, Payback and the Human Wage Comparison

Robot cost per hour is the total yearly cost of owning and running a robot divided by the hours it actually works. Eight lines make up that total: capital recovery, integration, maintenance, battery replacement, electricity, human supervision, software and insurance. Six of them are fixed annual costs, so the hours you run decide most of the answer. The other two, electricity and supervision, are paid on every hour, and when they already exceed the wage of the person the robot replaces, the robot never pays back at any utilisation.

Most robot business cases are built in a spreadsheet that starts with the purchase price and ends with a payback period in months. The purchase price is usually the least uncertain number in it. The numbers that decide the result are less visible: how many robots one person can supervise, how many hours a year the machine is actually working, how much of each hour it spends charging, how much output it produces compared with the person it replaces, and what maintenance costs once the warranty ends.

This guide works through each line with the arithmetic shown, across humanoids, quadrupeds, drones, arms and cobots, and AMRs. The equations are the same ones behind the Robot Economics Calculator, which lets you change every input and see the cost per hour, break-even utilisation and payback move.

The take: build the business case from cost per operating hour and compare it with the loaded wage times the robot's productivity ratio. Check the variable cost first: if electricity plus supervision per robot hour is above the displaced wage, stop, because no amount of utilisation fixes it. For humanoids today, the supervision ratio is the input that moves the answer most. For arms and AMRs it is utilisation and integration. Electricity is almost never the problem, and a three-year model that includes a warranty year understates maintenance.

Companion reading: robot power and batteries, robot teleoperation, how to choose a humanoid robot and warehouse and logistics robotics.

Table of contents

1. Key takeaways
2. The cost per hour equation
3. Capital recovery and residual value
4. Maintenance and the warranty trap
5. Batteries: replacement by cycles
6. Electricity: the smallest line
7. Supervision and teleoperation ratios
8. Integration, software and insurance
9. Utilisation is the biggest lever
10. Endurance, charging and fleet sizing
11. The human side: loaded wage and productivity
12. Break-even and the robot that never pays back
13. Simple payback and discounted payback
14. Worked examples by form factor
15. Business case checklist
16. Frequently asked questions
17. Sources

The cost per hour equation

Write H for annual operating hours. The total cost per operating hour is:

capital      = capital_annual / H
integration  = integration_annual / H
maintenance  = capex x maintenance_share / H
battery      = pack replacement cost per year / H
energy       = average_kW x price_per_kWh / charge_efficiency
supervision  = supervisor_cost_per_hour / robots_per_supervisor
software     = software_per_year / H
insurance    = insurance_per_year / H

total = sum of the eight lines

Two properties of this equation drive everything that follows.

First, six lines shrink as H grows, because they are annual sums spread over more hours. That is why utilisation matters so much: a robot that works 4,000 hours a year has half the capital, maintenance, software and insurance cost per hour of the same robot working 2,000 hours.

Second, two lines do not shrink at all. Electricity and supervision are paid on every hour the robot runs, so they set a floor under the cost per hour. Running the robot more never takes the total below that floor. If the floor is already above the human cost the robot is meant to replace, the case is lost before capital enters the picture.

This structure applies to every form factor. What changes between a cobot and a humanoid is which lines are large.

Capital recovery and residual value

Dividing the purchase price by the service life understates capital cost, because it ignores the cost of the money tied up in the machine. The standard engineering economics approach converts the purchase price into an equal annual payment using the capital recovery factor (CRF), and credits the resale value at the end through the sinking fund factor (SFF):

r   = annual interest or discount rate
n   = service life in years
CRF = r (1 + r)^n / ((1 + r)^n - 1)
SFF = r / ((1 + r)^n - 1)

capital_annual = capex x CRF - residual_value x SFF

The sinking fund factor is the annual amount that would compound to the residual value by the end of the life, so subtracting it gives credit for the resale today. When r is zero both factors reduce to 1/n.

Worked values at 8 percent:

Service life	CRF	SFF	Annual capital cost of 100,000 USD with 10% residual
3 years	0.3880	0.3080	35,723 USD
5 years	0.2505	0.1705	23,341 USD
7 years	0.1921	0.1121	18,087 USD
10 years	0.1490	0.0690	14,213 USD

The 5-year row works like this: 100,000 x 0.2505 = 25,046, minus 10,000 x 0.1705 = 1,705, gives 23,341 USD a year. At 2,000 operating hours a year that is 11.67 USD an hour of capital cost alone.

Three practical points:

• Service life is a judgement. For industrial arms there is evidence: the International Federation of Robotics models its installed base on a 12-year average service life and cites a pilot study that put the real figure closer to 15 years. For humanoids there is no fleet history at all, so a 3 to 5 year life is a planning assumption. Shorter lives raise capital cost sharply, as the table shows.
• Integration carries no residual. Commissioning, safety guarding, fixtures and software integration have no resale value. Recover them with the CRF alone.
• Use your real cost of capital. A leasing quote or a corporate hurdle rate is better than a generic 8 percent. The calculator lets you set it.

Maintenance and the warranty trap

Maintenance is usually modelled as a percentage of capex per year after the warranty ends. There is no audited cross-industry dataset for robot maintenance costs, so every figure in this table is a planning range. They are the ranges the calculator uses, and the confidence varies sharply by form factor.

Form factor	Planning range, % of capex per year	Confidence
Industrial arm or cobot	3 to 12, typical 6	Best supported. Trade and service contract figures cluster here
AMR	6 to 12, typical 8	Trade quotes. Budget pack replacement separately
Drone	10 to 25, typical 18	Trade quotes. Props, motors, gimbals and packs dominate
Quadruped	8 to 25, typical 15	Rule of thumb. Legged platforms wear joints, feet and cabling
Humanoid	15 to 35, typical 22	Engineering estimate. No mature commercial fleet exists to measure

The humanoid figure is reasoned from the hardware: dozens of actuators against six on a typical arm, hands as a high-wear part, and a battery in the torso. Treat any humanoid business case that assumes a low single-digit maintenance percentage as an assumption that no fleet has yet demonstrated.

The warranty trap. Warranties are short. Unitree lists an 8-month warranty for the standard G1 humanoid and 18 months for the G1 EDU. A three-year model that includes a warranty year averages maintenance over one free year and two paid years. On a 100,000 USD humanoid at 22 percent, the steady-state cost is 22,000 USD a year. The three-year average with a free first year is (0 + 22,000 + 22,000) / 3 = 14,667 USD, a third lower. The robot still costs 22,000 USD a year from year two onward, and a model built on the three-year average will look wrong in year four.

Model maintenance at the post-warranty rate for every year, and treat the warranty as a one-off saving in year one if you want to show it.

Batteries: replacement by cycles

Battery packs wear out by charge cycles, so count cycles rather than years. Cycle life is defined as the number of full cycles until the pack holds 80 percent of its original capacity. The calculation:

cycles_per_year = annual_operating_hours / runtime_per_charge
packs_per_year  = cycles_per_year / rated_cycles
battery_annual  = packs_per_year x pack_cost

Manufacturer figures anchor the cycle side:

Robot	Pack	Rated cycles	Source
MiR250 AMR	Lithium-ion, 13 h at max payload	Minimum 3,000 full charge cycles	MiR specifications
DJI Matrice 400 drone	TB100, 977 Wh	400 cycles	DJI specifications

For humanoids and quadrupeds that use NMC-type cells for energy density, vendors do not publish cycle ratings for the pack. The calculator uses 1,800 cycles as an engineering estimate for NMC and 4,000 for LFP; replace those with your vendor's figure when you have it.

Two worked cases show the spread:

• AMR. A MiR250 running 4,000 hours a year at 13 hours per full charge uses about 308 equivalent full cycles a year. At 3,000 cycles that is 9.75 years of pack life, longer than a typical 7-year depreciation period. Battery cost per hour rounds to a few cents.
• Drone. The Matrice 400 hovers for 53 minutes on one TB100. Keeping a 20 percent reserve leaves about 0.71 hours of usable flight per cycle (this reserve is a common operating practice, and the 20 percent figure is our assumption). At 400 flight hours a year that is about 563 cycles, which is 1.4 packs worn out a year. A drone flown daily needs new packs every year.

Pack price. BloombergNEF's 2025 survey put the average lithium-ion pack at a record low of 108 USD per kWh. That figure is dominated by electric vehicle packs bought in very large volumes. Robot packs are small, made in low volume, and carry their own BMS, enclosure, thermal management and often hot-swap connectors. The calculator uses 400 USD per kWh as an engineering estimate. Get a quote; the error in the business case from using the EV figure is usually small in dollars per hour, but it compounds on drones, where packs are replaced yearly.

Electricity: the smallest line

Energy cost per hour is average power draw times the electricity price, divided by charge efficiency. The US Energy Information Administration reports an average industrial electricity price of 9.64 cents per kWh in August 2026 (preliminary), and 14.47 cents for commercial customers. Using the industrial figure and a 90 percent charge efficiency for battery robots:

Robot	Average draw	Basis	Electricity per hour
UR10e cobot (mains)	350 W	Universal Robots: approx. 350 W on a typical program	3.4 cents
Spot quadruped	376 W	564 Wh pack over a 90 minute average runtime	4.0 cents
Humanoid	500 W	Engineering estimate	5.4 cents
Matrice 400 drone	1,106 W	977 Wh over 53 minutes of hover	11.8 cents

Against a loaded wage near 29 USD an hour, every one of these is under half a percent. Electricity matters in two narrower cases: robots with very high draw, such as a heavy industrial arm, and countries with very high business tariffs. Even then it rarely decides the case.

Electricity does matter for charging infrastructure and site power capacity, which are capital costs and belong in integration.

Supervision and teleoperation ratios

Supervision cost per robot hour is the loaded hourly cost of the supervisor divided by the number of robots one person covers:

supervision = supervisor_cost_per_hour / robots_per_supervisor

For the supervisor cost, BLS reports a May 2025 mean wage of 32.91 USD an hour for industrial machinery mechanics (SOC 49-9041). Employer costs for benefits push the total higher: in the BLS Employer Costs for Employee Compensation release for June 2026, wages and salaries were 70.0 percent of private industry compensation costs, which implies a multiplier of 1 / 0.70 = 1.43. That gives a loaded technician cost of about 47.01 USD an hour.

Robots per supervisor	Supervision per robot hour
1	47.01 USD
2	23.51 USD
5	9.40 USD
10	4.70 USD
15	3.13 USD

The ratio depends on whether the person is driving or watching:

• Teleoperation is one to one. Dexterous manipulation by a remote operator and humanoid demonstration data collection need a person for every robot hour. The robot then saves nothing on labour, because a person is still working every hour it works.
• Supervisory control means an operator watches several robots and takes over on exceptions. The achievable ratio depends on how often the robots need help. Humanoid and quadruped operators do not publish intervention rates, so any ratio for them is an assumption. The calculator's planning defaults are 1 to 2 for humanoids and 1 to 4 for quadrupeds.
• Arms and warehouse AMR fleets are covered by technicians responsible for a line or a fleet, so the ratio is really cells or robots per technician. The calculator's planning range is 5 to 40, with 15 as typical.
• Drones have one regulatory anchor. The FAA's proposed Part 108 rule for beyond visual line of sight operations, published 7 August 2025, would set a ratio of one unmanned aircraft per flight coordinator, except under a method acceptable to the FAA. The proposal leaves higher ratios to future consensus standards, manufacturer operating instructions and case-by-case review, and notes that current one-to-many operations have mostly been package delivery under Parts 107 and 135.

The supervision line is where humanoid business cases are usually won or lost. At a 1 to 2 ratio, supervision alone is 23.51 USD per robot hour, which is already above the productivity-adjusted wage in the first humanoid example below.

For how remote operation works in practice, see the robot teleoperation guide.

Integration, software and insurance

These three lines are fixed annual costs and are easy to leave out.

Integration covers site survey, mapping, guarding and safety assessment, fixtures and end effectors, connection to warehouse or manufacturing systems, and commissioning. For arms it is often the largest cost in the project. The calculator's planning defaults are 60 percent of capex for an arm, 40 percent for an AMR, 20 percent for a humanoid, 15 percent for a quadruped and 10 percent for a drone. These are planning estimates; a real quote replaces them. Recover integration over the service life with the CRF and give it no residual value.

Software includes fleet management, autonomy licences, cloud data, map updates and support subscriptions. Many vendors now price these per robot per year. Put them in at the quoted figure and check whether the price rises after an introductory term.

Insurance covers property damage to the robot and liability for what it does. The calculator uses 1 to 3 percent of capex a year as a planning range, highest for drones. Ask your broker; drone operations in particular often need specific aviation cover.

Utilisation is the biggest lever

Because six of the eight lines are annual costs divided by H, the hours a robot works each year move the cost per hour more than any other input except, for humanoids, the supervision ratio.

Take the humanoid assumptions from the worked examples (100,000 USD capex, 5-year life, 8 percent, 10 percent residual, 22 percent maintenance, 20,000 USD integration, 6,000 USD software, 2,000 USD insurance, 1 supervisor per 10 robots). Fixed annual cost is about 58,600 USD and the per-hour floor from electricity and supervision is 4.75 USD.

Annual operating hours	Pattern	Fixed cost per hour	Total per hour
1,000	Part-time pilot	58.63	63.38
2,000	One shift, five days	29.31	34.07
4,000	Two shifts	14.66	19.41
6,000	Three shifts, most days	9.77	14.52

(The 4,000-hour row here holds battery cost at the 2,000-hour rate for comparison; the full worked example in section 14 adds the extra pack wear.)

The total per hour falls fast at first and then flattens toward the 4.75 USD floor. This is why robots that can work round the clock, such as AMRs and arms on multi-shift lines, have the strongest economics, and why a robot bought for one shift is often under-used.

Annual operating hours should be realistic. A shift position is 2,000 hours a year (8 hours, 5 days, 50 weeks). Round-the-clock coverage is 8,760 hours. Subtract planned maintenance, breakdowns, shift changes, and for drones, weather and transit. Drones often fly only a few hundred hours a year; the calculator's default is 400.

Endurance, charging and fleet sizing

A battery robot does not work every hour it is on shift. Charging, swapping and returning to the dock take time, and that time decides how many robots you need to cover a position.

u               = runtime / (runtime + downtime)
working_min_hr  = 60 x u
robots_needed   = ceil(coverage_hours / (u x annual_hours_per_robot))

Downtime with no spare packs is the full recharge time. With spare packs, the robot only waits for the swap, provided the charger can refill a spent pack before the robot needs it again.

Manufacturer specifications give real inputs:

Robot	Runtime	Recharge	Utilisation without spares	Working minutes per hour
MiR250 AMR	13 h at max payload	Charging ratio up to 1:16 (10 min gives 2 h 40 min at max payload)	16 / 17 = 0.94	56.5
Spot quadruped	90 min average	60 min	1.5 / 2.5 = 0.60	36
Matrice 400 drone	53 min hover	45 min fast charge (220 V)	53 / 98 = 0.54	32
Unitree G1 humanoid	About 2 h	Not published	Depends on charge time	Depends on swap

Three patterns follow:

• Opportunity charging is how AMRs avoid the problem. With a 1:16 charge ratio, short top-ups at idle moments keep utilisation near 94 percent with no spare packs.
• Hot swapping is how quadrupeds and drones reach high utilisation. Spot without a spare works 36 minutes in each hour. With one spare pack and a short swap, the 60-minute recharge finishes well inside the 90-minute runtime, so utilisation approaches 100 percent minus the swap time. The same holds for the Matrice 400: its 45-minute fast charge is shorter than the 53-minute hover time, so one spare pack per aircraft keeps up if the charging station has a free channel. The Unitree G1 lists a quick-release battery, which makes swapping possible on that humanoid.
• Internal packs without swap force a full recharge. A humanoid with an internal pack, a 2-hour runtime and an assumed 90-minute recharge works about 34 minutes an hour (2 / 3.5 x 60). Covering one 2,000-hour shift position then takes two robots.

Fleet sizing matters for cost because coverage is what the customer pays for. If two robots are needed to cover one position, the cost per position hour is the cost per robot hour times two, divided by the utilisation-adjusted hours each supplies. The calculator computes robots per single-shift position and per round-the-clock position from these inputs.

The robot power and batteries guide covers pack chemistry, charging and swap design in more depth.

The human side: loaded wage and productivity

The comparison is with the full cost of the human hour the robot replaces.

Loaded wage. BLS reports a May 2025 mean hourly wage of 20.32 USD for laborers and freight, stock and material movers, hand (SOC 53-7062), with a median of 19.35 USD. Applying the 1.43 multiplier from the BLS compensation release gives a loaded cost of about 29.03 USD an hour. Use the local equivalent for your site, including agency margins, shift premiums, overtime, recruitment and turnover if you know them.

Productivity ratio. A robot rarely produces exactly what a person produces in an hour. Write p for the number of people one robot replaces for each hour it works:

displaced_cost_per_robot_hour = loaded_wage x p

A robot that works at 70 percent of a person's pace has p = 0.7 and displaces 0.7 x 29.03 = 20.32 USD an hour. An arm on a two-shift cell that replaces one operator per shift has p = 1 for each hour it runs. A drone that inspects a structure faster than a rope-access crew may have p above 1, measured against crew hours per task.

Early deployments usually run below p = 1. Measure p on a pilot, in units per hour, and use that measured figure. A business case that assumes p = 1 for a humanoid on day one is optimistic.

Break-even and the robot that never pays back

Split the cost per hour into fixed annual cost and variable cost per hour:

fixed_annual    = capital + integration + maintenance + battery + software + insurance (all per year)
variable_hourly = energy + supervision
margin          = loaded_wage x p - variable_hourly

if margin > 0:  break_even_hours = fixed_annual / margin
if margin <= 0: no break-even at any utilisation

The never-pays-back case. When the variable cost per hour is greater than or equal to the displaced cost, every extra hour of running loses money, so running more cannot recover the fixed cost. There is no break-even point. The calculator reports this case as having no break-even.

The common ways into this case:

• A tight supervision ratio. At 1 to 2, supervision is 23.51 USD per robot hour. A humanoid with p = 0.7 displaces 20.32 USD. The margin is negative before capital is counted.
• Low local wages. Where the loaded wage for the displaced job is a few dollars an hour and the supervisor is paid a skilled technician rate, the variable cost can exceed the wage even at modest ratios.
• Low productivity. p = 0.5 halves the displaced cost while supervision stays the same.

Cheap hardware does not escape this. A Unitree G1 lists at 13,500 USD before tax and shipping. Spread over 3 years and 2,000 hours a year, its capital cost is about 2.41 USD an hour. Supervised at 1 to 2, it still costs 23.51 USD an hour in supervision, and if it works at half a person's pace it displaces only 14.51 USD. It never pays back on that ratio. The hardware price barely matters when supervision is the dominant cost.

When there is a break-even. With a 1 to 10 ratio and the same humanoid assumptions, the margin is 20.32 minus 4.75, or 15.57 USD an hour. Break-even is 58,628 / 15.57 = 3,767 hours a year. Below that the robot costs more than the work it does. Above it, every extra hour earns 15.57 USD. A one-shift deployment at 2,000 hours is well below break-even, which is why the same robot can be uneconomic on one shift and economic on two.

If the break-even hours are above 8,760, the robot cannot reach break-even within a calendar year, and the case fails in practice even though the arithmetic gives a number.

Simple payback and discounted payback

Simple payback is the convention most vendors quote:

upfront         = capex + integration
opex_per_hour   = total - capital - integration
annual_net_cash = (loaded_wage x p - opex_per_hour) x H
payback_years   = upfront / annual_net_cash      (only when annual_net_cash > 0)

The capital recovery lines are removed from the running cost here, because the upfront cash is counted directly.

Simple payback ignores the time value of money. Discounted payback asks how many years of savings, each discounted back to today, it takes to recover the upfront cost. With equal annual savings A, upfront cost P and discount rate r:

discounted_payback_years = -ln(1 - P x r / A) / ln(1 + r)     (only when P x r < A)

Two examples at r = 8 percent:

Upfront	Annual net cash	Simple payback	Discounted payback
120,000 USD	66,539 USD	1.80 years	2.02 years
120,000 USD	25,000 USD	4.80 years	6.30 years

For short paybacks the two are close. For long paybacks the gap is large: in the second row, the discounted payback of 6.3 years is beyond a 5-year service life, so the robot does not recover its cost in present-value terms even though simple payback says 4.8 years. When simple payback is more than about half the service life, compute the discounted figure.

Payback also hides the case where a positive annual saving is too small to matter. In the 1 to 10 humanoid example at 2,000 hours, opex per hour is 19.89 USD against a displaced 20.32 USD, so annual net cash is about 852 USD and simple payback is about 141 years. A payback period longer than the robot's life is equivalent to no payback.

Worked examples by form factor

All examples use the same labour and energy inputs: a displaced loaded wage of 29.03 USD an hour (BLS SOC 53-7062 mean times 1.43), a supervisor at 47.01 USD an hour (BLS SOC 49-9041 mean times 1.43), electricity at 9.64 cents per kWh, an 8 percent rate and a 10 percent residual. Capex, life, hours, maintenance, integration, software, insurance, supervision ratio and productivity are planning assumptions from the calculator's defaults, stated in each row. They are illustrations to show how the lines interact. Use your own quotes.

Assumptions

Case	Capex	Life	Hours/yr	Maint.	Integration	Software, insurance per yr	Ratio	p
Humanoid A	100,000	5	2,000	22%	20,000	6,000, 2,000	1:2	0.7
Humanoid B	100,000	5	2,000	22%	20,000	6,000, 2,000	1:10	0.7
Humanoid C	100,000	5	4,000	22%	20,000	6,000, 2,000	1:10	1.0
Quadruped	75,000	5	1,500	15%	11,250	4,000, 1,125	1:4	0.5
AMR	50,000	7	4,000	8%	20,000	3,000, 750	1:15	1.0
Cobot arm	45,000	10	4,000	6%	27,000	1,500, 450	1:15	1.0

Battery inputs: humanoid 2.25 kWh pack (the Figure 02 size) at 900 USD, 3.6 h runtime and 1,800 cycles (estimates); quadruped 564 Wh pack at 226 USD, 1.5 h runtime and 1,800 cycles; AMR pack at 600 USD (estimate), 13 h runtime and 3,000 cycles; the arm is mains powered.

Cost per operating hour, USD

Line	Humanoid A	Humanoid B	Humanoid C	Quadruped	AMR	Cobot arm
Capital	11.67	11.67	5.84	11.67	2.26	1.60
Integration	2.50	2.50	1.25	1.88	0.96	1.01
Maintenance	11.00	11.00	5.50	7.50	1.00	0.68
Battery	0.14	0.14	0.14	0.08	0.02	0.00
Electricity	0.05	0.05	0.05	0.04	0.03	0.03
Supervision	23.51	4.70	4.70	11.75	3.13	3.13
Software	3.00	3.00	1.50	2.67	0.75	0.38
Insurance	1.00	1.00	0.50	0.75	0.19	0.11
Total	52.87	34.07	19.48	36.34	8.34	6.94
Displaced cost (wage x p)	20.32	20.32	29.03	14.51	29.03	29.03
Break-even hours	None	3,767	2,427	13,536	800	583
Simple payback	None	About 141 years	1.8 years	None in practice	0.73 years	0.73 years

What the table shows:

• Humanoid A never pays back. Supervision at 1 to 2 (23.51) exceeds the displaced 20.32 on its own. Supervision is 44 percent of the total.
• Humanoid B has a break-even at 3,767 hours but runs only 2,000, so it is still uneconomic.
• Humanoid C shows what has to be true for a humanoid to work: two shifts, a 1 to 10 ratio and full human productivity. Then cost per hour falls to 19.48 USD and simple payback is 1.8 years (2.0 years discounted). Each of those three assumptions is unproven for current humanoid fleets.
• The quadruped has a break-even of 13,536 hours, above the 8,760 hours in a year. Inspection quadrupeds are usually justified on safety, data quality and access to hazardous areas, which this wage comparison does not capture.
• The AMR and cobot arm look very strong, with paybacks under a year. That result depends on each robot fully replacing one person for every hour it runs at a 1 to 15 technician ratio. Real projects often see lower p, more downtime and higher integration, so treat these as upper bounds and run your own pilot data through the calculator.

Drones are compared with crew hours per task rather than a warehouse wage, so they are left out of the payback table. On the cost side, a Matrice 400 class aircraft at an assumed 12,000 USD capex, 4-year life, 400 flight hours a year, 18 percent maintenance and a 1 to 1 pilot ratio costs about 70 USD per flight hour, of which 47 USD is the pilot. Moving to one pilot for several aircraft, where the regulator allows it, is the largest lever for drone economics.

To run any of these cases with your own numbers, open the Robot Economics Calculator, pick the form factor and change the inputs. It shows the eight-line breakdown, break-even utilisation, payback and a sensitivity chart of which input moves the total most.

Business case checklist

1. Measure productivity on a pilot. Record units per hour for the robot and for people doing the same task. Use the measured ratio p.
2. Check the variable floor first. Electricity plus supervision per robot hour must be below loaded wage times p. If it is not, stop and work on the supervision ratio.
3. Use the post-warranty maintenance rate for every year, and show the warranty as a separate one-off saving.
4. Count battery cycles. Divide annual hours by runtime per charge, compare with the rated cycles, and budget pack replacements as step costs.
5. Set realistic operating hours. Start from the shift pattern, then subtract charging, maintenance, breakdowns and, for drones, weather.
6. Size the fleet for coverage. Use utilisation to work out how many robots one position needs, including spare packs and chargers.
7. Include integration, software and insurance at quoted figures. Recover integration with no residual value.
8. Recover capital with the CRF at your real cost of capital and a defensible residual value.
9. Report discounted payback whenever simple payback exceeds half the service life.
10. Run a sensitivity. Vary each input by 30 percent and see which one moves the total most. For humanoids it is usually the supervision ratio; for arms and AMRs, utilisation or integration.

For fleet-level operating data that feeds these inputs, see the robot fleet management guide. For selection criteria by form factor, see how to choose a cobot and how to choose an AMR or AGV.

Frequently asked questions

How do you calculate a robot's cost per hour? Add eight annual or hourly costs and express each per operating hour: capital recovery, integration, maintenance, battery replacement, electricity, supervision, software and insurance. Six are annual amounts divided by annual operating hours. Electricity and supervision are paid per hour. Convert the purchase price to an annual cost with the capital recovery factor, crediting the residual value through the sinking fund factor.

What does a humanoid robot cost per operating hour? On planning assumptions of 100,000 USD capex, a 5-year life, 2,000 hours a year, 22 percent annual maintenance and US labour costs, a humanoid costs about 34 USD an hour with one supervisor per 10 robots and about 53 USD an hour with one per 2. Supervision and maintenance dominate. Electricity is about 5 cents an hour.

How much does electricity add to a robot's running cost? Very little. At the August 2026 US industrial average of 9.64 cents per kWh, a 350 W cobot costs about 3.4 cents an hour, a 500 W humanoid about 5.4 cents and a drone drawing about 1.1 kW in hover about 12 cents. That is well under 1 percent of a loaded wage near 29 USD an hour.

How many robots can one person supervise? True teleoperation is one to one. Supervisory control of humanoids and quadrupeds has no published industry data, so planning ratios of 1 to 2 and 1 to 4 are assumptions. Technicians covering arms and AMR fleets look after many units each, with 15 as a common planning figure. For drones, the FAA's proposed Part 108 would default to one aircraft per flight coordinator unless the FAA accepts a method for more.

When does a robot never pay back? When electricity plus supervision per robot hour is equal to or above the loaded wage of the person it replaces multiplied by the productivity ratio. Each extra hour then loses money, so higher utilisation cannot recover the fixed costs. Tight supervision ratios, low local wages and low productivity are the usual causes.

How do I calculate break-even utilisation for a robot? Divide the fixed annual cost (capital, integration, maintenance, battery, software and insurance) by the hourly margin, which is the displaced wage times productivity minus electricity and supervision per hour. The result is the annual operating hours at which the robot costs the same as the labour it replaces. If it exceeds 8,760 hours, the robot cannot break even in practice.

Why do three-year TCO models flatter robots? Because the first year is usually under warranty, so maintenance in that year is close to zero. Averaging one free year with two paid years understates the steady-state maintenance cost by a third, and the error shows up from year four onward. Model the post-warranty rate for every year.

How long do robot batteries last? Count cycles. Divide annual operating hours by runtime per charge to get cycles per year, then divide the rated cycle life by that. A MiR250 AMR rated for at least 3,000 cycles and running 4,000 hours a year at 13 hours per charge lasts about 9.75 years. A DJI Matrice 400 pack rated for 400 cycles and flown 400 hours a year wears out in under a year.

Sources

Primary sources first: government statistics, manufacturer specifications and the regulator's proposed rule. The planning ranges for maintenance, supervision ratios, integration, software, insurance and humanoid power draw are engineering estimates from the Robo2u calculator and are labelled as such in the text.

Primary

1. BLS Employer Costs for Employee Compensation, June 2026 - private industry wages and salaries at 70.0 percent of total compensation costs, the basis of the 1.43 multiplier.
2. BLS Occupational Employment and Wage Statistics, May 2025, SOC 53-7062 - laborers and freight, stock and material movers, hand: mean 20.32 USD and median 19.35 USD an hour (retrieved via the BLS public data API, series OEUN000000000000053706203 and 08).
3. BLS Occupational Employment and Wage Statistics, May 2025, SOC 49-9041 - industrial machinery mechanics: mean 32.91 USD an hour (series OEUN000000000000049904103).
4. EIA Electric Power Monthly, Table 5.6.A - average US industrial price of 9.64 cents per kWh and commercial 14.47 cents in August 2026 (preliminary).
5. FAA and TSA, Normalizing Unmanned Aircraft Systems Beyond Visual Line of Sight Operations, proposed rule - proposed Sec. 108.210 on multiple aircraft, the 1:1 default ratio and case-by-case review.
6. IFR World Robotics 2017, Chapter 1 - 12-year average service life assumption and the pilot study pointing to about 15 years.
7. MiR250 specifications - up to 13 h at maximum payload, charging ratio up to 1:16, minimum 3,000 full charge cycles.
8. DJI Matrice 400 specifications - TB100 977 Wh, 400 cycle count, 59 minute flight and 53 minute hover times, 45 minute fast charge at 220 V.
9. Boston Dynamics Spot - 564 Wh battery, 90 minute average runtime, 60 minute recharge.
10. Universal Robots UR10e technical specifications - power consumption approx. 350 W using a typical program.
11. Unitree G1 - 13,500 USD list price excluding tax and shipping, about 2 h battery life, quick-release battery, 8 and 18 month warranties.
12. Figure unveils Figure 02 - 2.25 kWh battery pack in the torso.
13. BloombergNEF: lithium-ion battery pack prices fall to 108 USD per kWh - 2025 battery price survey, 9 December 2025.

Tool

14. Robot Economics Calculator - the equations in this guide, the planning defaults by form factor with confidence labels, and the sensitivity chart.