Mobile Robots
Warehouse Robot Throughput: Do the Math Before the Pilot
How to calculate mobile robot throughput from travel distance, dwell time and congestion, with a worked example showing why the second robot never doubles the rate.
Throughput per robot equals 3,600 divided by the full round-trip time, and the round trip is dominated by dwell, not by driving. A typical goods-to-person robot spends 40 % to 60 % of its cycle stationary at pick stations and charging points, which is why doubling travel speed raises throughput by roughly 15 % to 25 % rather than by half.
Building the round trip
Break every trip into six terms and measure or estimate each.
- Travel to source. Distance divided by effective speed. Effective speed is lower than rated speed because of acceleration, turns and safety slowdowns; use 60 % to 75 % of rated as a first estimate.
- Docking and load. 5 s to 30 s depending on whether the robot lifts a shelf, receives from a conveyor or waits for a person.
- Travel to destination. As above.
- Dwell at the station. The largest single term in most goods-to-person systems, 15 s to 60 s while a picker works.
- Return or reposition. Often to a buffer rather than the origin.
- Charging amortised. Charge time divided by the number of trips per charge, added to every trip.
| Term | Value | Note |
|---|---|---|
| Travel to shelf, 42 m at 1.1 m/s effective | 38 s | Rated 1.6 m/s, 69 % effective |
| Lift shelf | 9 s | Align, jack up |
| Travel to station, 55 m | 50 s | Loaded, slightly slower |
| Dwell at pick station | 34 s | Picker takes 2.4 lines |
| Travel back, 55 m | 50 s | |
| Lower shelf | 7 s | |
| Charging amortised | 11 s | 18 min per 4 h of work |
| Round trip | 199 s | |
| Trips per hour per robot | 18.1 | 3,600 / 199 |
| Lines per hour per robot | 43.4 | at 2.4 lines per shelf visit |
Congestion is the term that ruins forecasts
Robots interfere with each other. Aisles, intersections, lift queues and station approaches all serialise traffic, and the loss grows non-linearly with fleet size. A useful planning rule from installed systems: expect no measurable loss below roughly 0.5 robots per 100 m² of drivable area, 10 % to 15 % loss around 1.0, and 25 % or more above 1.5.
Applying that to the example: at 18.1 trips per hour and 43.4 lines per hour per robot, a target of 1,200 lines per hour looks like 28 robots. With 15 % congestion loss it is 33. With station queuing at peak it is closer to 36. The difference between the naive and the realistic number is eight vehicles, which is a substantial budget error.
The levers, ranked by effect
| Lever | Typical gain | Cost |
|---|---|---|
| Reduce dwell by batching picks per visit | 15 to 40 % | Software and process |
| Add pick stations when dwell-bound | 10 to 30 % | Moderate capital |
| Slot fast movers nearer the stations | 10 to 25 % | Low, analysis only |
| Improve traffic rules and intersection logic | 5 to 20 % | Software |
| Opportunity charging instead of full charge cycles | 4 to 10 % | Charger count |
| Raise robot travel speed | 5 to 15 % | Often limited by safety distance |
| Add robots | linear then diminishing | High |
Slotting is the underrated one. Moving the top 20 % of SKUs closer to the stations shortens the two travel terms simultaneously, and it costs nothing but a data exercise and a weekend of relocation.
Designing a pilot that predicts anything
A three-robot pilot in an empty aisle tells you the vehicle works. It tells you nothing about congestion, charging strategy or station queuing, which are the three things that determine whether the full system meets its rate. A pilot worth running should include at least the peak robot density in a representative area, real order profiles rather than sample data, and a full charging cycle across a shift.
Frequently asked questions
How do I calculate mobile robot throughput?
Sum travel, docking, dwell, return and amortised charging into a round-trip time in seconds, then divide 3,600 by it for trips per hour per robot. Multiply by lines per trip, then subtract a congestion allowance.
Why does doubling robot speed barely help?
Because 40 % to 60 % of the cycle is stationary time at pick stations, docking and charging. Halving the travel terms only touches the other half, so the realistic gain is 15 % to 25 %, and safety-limited speeds often prevent even that.
How much throughput does congestion cost?
Roughly nothing below 0.5 robots per 100 m² of drivable area, 10 % to 15 % near 1.0, and 25 % or more above 1.5. The loss grows faster than the fleet, so it must be modelled rather than assumed away.
Should I add robots or pick stations?
If dwell time is the largest term in the round trip, add stations. Robots then queue less and each one completes more trips, which usually delivers more lines per hour per euro than additional vehicles.
What makes a pilot representative?
Peak robot density in the test area, real order profiles rather than samples, and at least one full charging cycle across a shift. Three robots in an empty aisle prove the vehicle works and predict nothing about the system.
Sources
- ISO 3691-4, safety requirements for driverless industrial trucksInternational Organization for Standardization, speed and detection field requirements that cap travel speed
- World Robotics, service robots report seriesInternational Federation of Robotics, logistics robot deployment data
- Robotics at NISTNational Institute of Standards and Technology, robot performance measurement and test methods