AMR fleet charging: the fleet stops before the battery does

Whoever runs a shift notices it at the same moment every time: mid-afternoon, missions queued, and two vehicles parked next to the charging station. Neither is really flat, but the third one to arrive finds the spot taken and waits, and meanwhile the pallet round that should have closed before shift change slips. The question the manager usually takes to the supplier is about battery runtime. It is almost always the wrong question, because fleet availability is decided by the geometry of the stops and the number of charging points well before the watt-hours in the pack.

Three numbers, and the rest is arithmetic

Sizing starts from energy, not power, and it needs three quantities the site already knows: the usable capacity of the pack, the energy one typical mission consumes, and the minutes actually available at a station during the busiest shift. The average power required is the ratio between the energy to put back and the window in which it can be put back. Take an example with our own numbers, chosen to show the method. A 3 kW in-line charging point, in a real six-minute stop, returns 0.3 kWh. If the average round consumes 0.25, the balance breaks even on every cycle. If it consumes 0.6, the fleet loses half of what it needs on every round, and the drift stays invisible until the state of charge drops below threshold in the worst hour.

The second piece of arithmetic is utilization, and it is the one that takes apart most over-optimistic projects. A vehicle running at 95% utilization on an eight-hour shift has 24 minutes of downtime in total, broken into one- or two-minute fragments scattered wherever they fall. No charging strategy survives that distribution, however fast the charger. Adding one vehicle to the fleet lowers everyone's utilization and creates the window. It costs less than it seems, because the extra vehicle is there mainly to let the others charge.

Three strategies, three different ways of losing vehicles

The workable strategies are few, and the choice depends on where the gaps in the cycle are. Opportunity charging uses the natural stops, many and short, and keeps vehicles in the round. Scheduled charging concentrates charging into one long session: it is simpler to manage, but it takes the vehicle out of rotation for a block of time. Battery swapping cuts downtime to a few minutes at the cost of spare batteries, a dedicated space and a manual procedure someone has to carry out. The first model rewards chemistries that tolerate frequent partial charges, while old lead-acid pays for them with sulfation, as a review published in Robotics Tomorrow in August 2026 by an intralogistics systems integrator points out.

The same text contains the methodological advice worth more than any figure. Before choosing, simulate the state of charge of the whole fleet across the busiest shift, at the real duty cycle rather than the average one. Sizing done on the average works eleven months a year and fails in the month that counts. The method for rebuilding the worst hour from movement history is described in our article on evaluating an autonomous forklift and AMR project.

The queue grows faster than the fleet

Published studies also suggest that the bottleneck is the station rather than the cell. In January 2023 Hu, He and Shen published in Frontiers in Neurorobotics a comparison of two ways of planning missions and charging. The simulated fleets are heterogeneous, with 5, 10 and 15 AGVs, in a warehouse with five charging stations. The integrated method, which decides missions and station stops together, cuts average completion time by 13.62% and average delay by 76.69% compared with the sequential method. The most instructive figure is how the gap grows. The difference in average delay between the two methods is 219 seconds with 50 missions and 5 vehicles, and rises to 3,591 seconds with 100 missions and 15 vehicles. The bigger the fleet and the workload, the more it matters how you decide who goes to charge and when. The two operating constraints the authors model are ones every shift supervisor knows: the vehicle must finish its current mission before going to charge, and it cannot show up at the charger with a load on board.

On threshold behavior, the most recent reference is a reinforcement learning simulation published by Shaji, Sobhanan and Defryn in July 2026, on scenarios with four and six vehicles and only two stations. The learned policy sends vehicles to charge when the state of charge falls between 24% and 28% and brings them back into service between 75% and 95%. That way it completes about 6% more orders than the best benchmark heuristic. The study's units are abstract and the values don't transfer as they are to a real site. The shape of the policy does: vehicles leave the charger well before 100%, because the last few percentage points cost station time and take it from the vehicle that is waiting.

Who decides when a vehicle leaves the round

The decision belongs to the fleet manager, and that is where to look when putting together a fleet of mixed brands. The VDA 5050 specification, which reached version 3.0.0 in March 2026, standardizes exactly that interface, over MQTT with JSON messages. For charging it defines the data, namely the stateOfCharge, batteryHealth, charging and range fields of the powerSupply object. It also defines the startCharging and stopCharging actions, available both at a charging spot with the vehicle stopped and on a charging lane traveled while moving. The text excludes safety, traffic management logic and cybersecurity from its scope, while it lists energy management among the functions of fleet control. In practice the specification makes charging observable and controllable in a uniform way, and leaves it to whoever runs the fleet to decide how many stations are needed and which thresholds to use.

One detail of the same specification deserves attention at purchase stage. The field that reports battery health is a percentage, but the method for calculating it sits in the manufacturer's BMS and not in the standard. Two fleets that both report 85% can be in very different condition. Anyone mixing vehicles from different suppliers should have each of them state, by contract, how they define that number and at what threshold they consider the pack at end of life.

Chemistry decides how often you can split the charge

For the two most common chemistries the reference data, taken at cell level rather than pack level, are still those summarized by Battery University. Lithium iron phosphate sits around 90-120 Wh/kg, with typical charging at 1C, more than 2,000 cycles and a thermal runaway onset around 270 °C. NMC rises to 150-220 Wh/kg, with 1,000-2,000 cycles and an onset around 210 °C. These are category ranges, useful as orders of magnitude more than as product specifications, and they are enough to steer the choice. Where the vehicle can carry the weight, the cycle life and thermal margin of the more robust chemistry pay off in a fleet that does dozens of partial charges a day. Where the pack has to fit a compact chassis, the density of the more energetic chemistry decides.

On the hidden cost of haste there is a quantitative figure, although it comes from a simulation. A battery-health-aware scheduling study published on arXiv in 2026 assigns fast charging, in its model, about three times the wear of standard charging. It also adds the aging of vehicles left idle at a high state of charge while waiting for a mission. The policy that accounts for both cuts overall degradation by more than half compared with rule-based dispatch, at the same throughput. The coefficients come from a model, without measurements on real cells, and they still show where the time gained with a more powerful charger ends up: in the life of the pack.

Below freezing, the question is whether charging is allowed

In refrigerated warehouses the constraint changes nature and concerns whether charging is possible at all. Below 0 °C lithium tends to deposit as metal on the anode instead of intercalating into the graphite, with permanent capacity loss and a fragility that later shows up under vibration. Battery University's practical guidance is clear: no charging below freezing, and at -30 °C the allowable current drops to one fiftieth of rated capacity, meaning charge times in the tens of hours. Some special cells accept current down to -10 °C, at a reduced rate and with a lower end-of-charge voltage, reaching about 80% of capacity in more than twelve hours.

The mechanism was measured by Chen and colleagues in a 2025 study in ACS Electrochemistry. Plating becomes possible when the potential of the graphite electrode drops to 0 V against lithium metal. The charge the electrode accumulates before getting there collapses with the cold: 412 mAh/g at -20 °C, 160 at -30 °C and 106 at -40 °C, measured at a very low current of about C/24. The first value exceeds graphite's theoretical intercalation capacity of 372 mAh/g, because it measures all the charge passed and not only the intercalated part. The measurements also come from a three-electrode laboratory cell, not from an industrial pack. They therefore hold as a mechanism rather than as system performance, but the operational consequence is the same. In a cold store there are two routes. Either the station moves outside the refrigerated zone, with the transfer time budgeted, or the fleet uses packs with BMS-managed built-in heating, which bring the cells up to temperature before accepting current. For comparison, with lead-acid batteries the cold costs capacity more than it imposes bans. A technical manager at a truck manufacturer, in a trade magazine in December 2024, put the loss at around 30% of rated capacity below freezing. It is an interested party's statement, useful as an order of magnitude.

Even with lithium, the charging area is still a technical room

The regulatory picture splits into separate families, and mixing them up leads to quotes that forget something. Machine safety is governed by EN ISO 3691-4:2023, which covers driverless trucks and their systems, charging stations included, and deals with their electrical and electronic requirements in clause 4.1.3. The electrical requirements of industrial trucks in general are in EN 1175, now in the 2025 edition that replaced the 2020 one. Its scope expressly excludes the functions used for the automatic operation of driverless trucks and refers to EN ISO 3691-4:2023 for them. The safety of industrial lithium cells and batteries, including truck traction applications, is in IEC 62619:2022. Its second edition added requirements on moving parts, hazardous live parts, battery system design and electromagnetic compatibility. The installation and operation of traction batteries, explicitly including those of automated guided vehicles, is in IEC 62485-3:2014, which contains the method for calculating ventilation during charging.

That calculation method comes from lead-acid gassing, which is why in many plants the charging area is a ventilated room with a sign on the door. With lithium the risk changes nature: less hydrogen to dilute, more thermal energy to contain and separate, with the same design rigor applied to a different problem. This is our own reading, with no direct backing in a standard, and it avoids the worst conversation, the one in which charging ends up wherever there was room. In compact automated warehouses the position of the stations also enters the fire protection design, as our piece on sprinkler criteria for automated warehouses explains.

The stated number and the figure that eats it

On the benefits of charging built into the process, the most quoted figure in Europe comes from a September 2024 consulting study released by the supplier of the inductive charging system. It speaks of "+50% productivity" for the AGV fleet, of "-30%" fleet size in an alternative scenario, of "more than 5,000 hours a year" of charging downtime avoided and of "10,900 avoided crossings" between people and vehicles. The source itself declares the conflict of interest, fleet size and number of shifts are not published, and without those two numbers none of the percentages can be checked. Next to that 50% belongs the figure that erodes it, the same one seen above. The strategy that keeps vehicles always charged is also the one that charges them fast, and fast charging costs pack life to a degree that is anything but negligible.

Then there is an effect that concerns people more than machines. When the station sits off the natural round, vehicles go there reluctantly and shift workers learn to put it off until the state of charge turns critical. At that point nearly all of them go to charge together, at the worst moment, and the queue everyone wanted to avoid forms anyway. Station placement, in other words, is an organizational choice before it is an engineering one.

What to send us for an assessment

Credible sizing starts from warehouse data the site already holds, even when nobody has ever pulled it together:

• missions per shift and their hourly distribution, with the busiest hour kept apart from the average;
• average distances per mission, weights handled and any slopes or ramps;
• number of shifts per day and length of breaks, which are the candidate windows;
• a layout with the points where vehicles already stop today and the power available there;
• operating temperatures, especially any refrigerated zones;
• expected vehicle utilization and battery chemistry, if the fleet already exists.

You can send them by email to info@dipitech.com or on WhatsApp. From these numbers come the pack capacity, the number and position of the stations, and the thresholds for entering and leaving charge. Almost always the decisive recommendation comes out too: if the average stop returns less energy than the round consumes, no battery solves the problem. The choice is then between moving the station onto the route, adding one or adding a vehicle. The available solutions are described on the pages about material handling automation and mobile robotics and robotic forklifts and AMRs for pallets. The overall business case is in our analysis of AMR ROI, TCO and payback, and phasing into a warehouse that is already running is the subject of our article on automating pallet handling in a running warehouse.