At a glance: A service robot battery is not an EV battery scaled down. The duty cycle is spiky, partial and opportunistically charged, and the specifications that decide whether a pack survives are the continuous and pulse discharge ratings, the charge-acceptance curve above 80 percent, and the communication interface the fleet can actually read. This guide covers the pack itself — chemistry by duty profile, a peak-load sizing method, what RS485 with a published register map buys an operator, and the five sourcing questions that decide whether a replacement pack exists in year four.
Ask a fleet operator what limits their service robots and you will rarely hear about navigation or payload. You will hear about batteries: how long a shift really runs, why one unit is down to 70 percent capacity at eighteen months while its twin is still at 88, and why the replacement pack quoted at year three costs more than the original. Search behaviour reflects that gap — queries like "battery for service robots", "service robot lithium battery pack" and "nmc battery for service robots with rs485" all appear in current search data, and the pages that rank for them are mostly general lithium-battery primers that say nothing about the specific duty cycle of an autonomous platform.
This guide covers the pack itself, not the charging architecture. It sets out the three specifications that actually determine lifetime and safety in a service robot — chemistry at the cell level, the continuous-discharge rating relative to duty cycle, and the communication interface between the pack and the robot controller — plus the sourcing questions that decide whether a replacement pack will be available in year four.
Why Service Robot Packs Fail Differently From EV Packs
A service robot battery is not an electric-vehicle battery scaled down, and treating it as one produces packs that fail early in a predictable way. The duty cycle is different in four respects that matter for cell selection.
| Property | Service robot (AMR / scrubber) | Passenger EV | Consequence for pack design |
|---|---|---|---|
| Operating pattern | Frequent, short partial cycles with 20–40 percent depth of discharge, punctuated by long low-power idle | Deep cycles, 60–90 percent depth of discharge | Cycle-life specification must be read at the partial-cycle depth actually used, not the 100 percent figure on the datasheet |
| Discharge profile | Spiky: brush motors and pump surges draw 2–4× the average load for 1–3 seconds | Smoother, motor inverter with managed ramp | Continuous and pulse discharge ratings matter more than energy density |
| Charge window | Opportunistic: 20–60 minute docks between tasks, often at high rate | Scheduled, typically 7–11 kW overnight | Charge acceptance at the top of the range is the limiting factor, not capacity |
| Thermal environment | Wide ambient range, no liquid cooling loop, pack enclosed in a moving chassis | Actively managed thermal system | Passive thermal design must be validated at the worst ambient case, not the nominal |
The spiky discharge profile is the one that most often goes unspecified. A delivery platform with a rated 150 W average draw will momentarily pull 600 W when it starts on an incline or operates a lift mechanism. If the pack's continuous rating is sized to the average, the battery protection board will trip on current limit under exactly the conditions the robot encounters a dozen times a shift. The symptom operators report is a unit that stops mid-task with a fault that clears on reboot — which reads as a software bug and is diagnosed as a battery-sizing problem only after weeks of investigation.
Chemistry: LFP, NMC and LTO Read Differently at Pack Level
The cell chemistry decision for a service robot is less about energy density than about cycle life at partial depth and behaviour at temperature extremes. The three options commonly specified behave as follows.
| Chemistry | Typical cycle life at 80% DoD | Energy density (Wh/kg, pack) | Cold behaviour | Where it fits |
|---|---|---|---|---|
| LFP (lithium iron phosphate) | 3,000–5,000 | 90–120 | Capacity dips sharply below 0 °C; usable to −10 °C with derating | Indoor cleaning and delivery fleets that run daily duty cycles and are charged indoors |
| NMC (nickel manganese cobalt) | 1,200–2,500 | 150–200 | Better than LFP to −20 °C; more usable cold capacity | Cold-storage, outdoor winter and long-shift platforms where pack mass is constrained |
| LTO (lithium titanate) | 15,000–25,000 | 60–80 | Excellent to −30 °C | High-rate opportunity charging, three-shift operations, freezer environments |
The commercially important line in that table is the cycle-life column, because it interacts with the partial-cycle pattern described above. An LFP pack operating at 30 percent depth of discharge on a well-managed charger will commonly deliver well beyond its nameplate cycle count, because deep-discharge stress is not occurring. An NMC pack in the same duty can be the better engineering choice on cold sites and the worse financial choice in a warm indoor facility, where it will reach end of life on calendar and cycle ageing before an LFP equivalent. The decision should be made against the actual duty profile — shift length, discharge depth, ambient range and charge-window length — rather than against a headline energy-density figure.
One further pack-level distinction is worth stating because it appears in procurement documents as a confusion: the cell format. Cylindrical 18650 and 21700 cells dominate service-robot packs because they are commodity items with multiple second sources, well-characterised thermal behaviour and replaceable cell-level service. Prismatic and pouch cells offer higher packing efficiency in the same volume, which matters when the pack envelope is fixed by an existing chassis, but they generally commit the buyer to a single cell supplier and a whole-pack replacement path. For a fleet expected to run five or more years, this is a sourcing question as much as an engineering one, and it is covered in the section below.
Discharge Rating: The Specification That Gets Skipped
Three ratings on a cell datasheet determine whether a pack will survive a service-robot duty cycle without nuisance trips and without accelerated ageing.
The arithmetic that ties these together is a peak-load budget: list every actuator on the platform, take the simultaneous worst case, add 20 percent margin, and compare the result to the pack's pulse rating at the maximum ambient temperature in the specification. If the number does not fit, the fix is a higher-capacity or higher-rate pack, not a software current limit — a software limit applied to a pack that trips at the hardware level simply moves the failure.
BMS and Communication: What RS485 Actually Buys You
The battery management system is where a service-robot pack differs most from a consumer pack, because the robot's fleet software needs to know the battery's state in enough detail to plan work. The specification that matters is the communication interface and what data moves across it.
| Interface | Typical use on a service robot | Data available | Trade-off |
|---|---|---|---|
| RS485 with a documented protocol (often Modbus RTU over RS485) | Wired link from pack to robot main controller; the standard in most commercial AMR and scrubber platforms | State of charge, voltage, current, cell group voltages, temperatures, cycle count, protection flags, firmware version | Requires the protocol register map to be published or the integrator has to reverse-engineer it |
| CAN bus | Higher-end platforms, multi-pack configurations, automotive-derived designs | As above, at higher update rates and with multi-node support | Robust and well suited to multi-pack, but adds harness and transceiver cost |
| SMBus / I2C to an internal controller | Smaller platforms, where the robot controller polls the pack directly | Reduced set, typically charge state and basic protection flags | Cheapest, but limits fleet-level battery analytics |
RS485 is the interface most fleet buyers should require, and the reason is operational rather than technical. With a documented register map, the fleet management layer can read per-pack state of charge and cycle count, which is what turns battery health from an anecdotal report into a managed asset. A fleet operator can then see that one pack in a group of thirty is cycling 2.4× faster than the others because that robot is assigned the longest routes, and rebalance routes before the pack reaches end of life early. Without a readable interface, the same information arrives only when the pack fails, and it arrives as an unplanned downtime event.
The specification to write into a purchase order is therefore: RS485 with a published register map, per-cell-group voltage reporting, at least two temperature sensors per pack, protection flags readable without a proprietary tool, and the ability to read cycle count and cumulative throughput. A vendor who cannot supply a register map is a vendor whose pack cannot be integrated into a fleet data layer, whatever its datasheet says.
State of health, not just state of charge
Two numbers matter and only one is usually available. State of charge tells the fleet how much work remains in this shift. State of health — the pack's present capacity as a percentage of its original rated capacity — tells the operator how much work the pack will be able to do in six months. A pack reporting 100 percent state of charge while its state of health has fallen to 74 percent will complete a shorter shift than it did when new, and the operator will read that as a navigation or scheduling problem. If the BMS does not report state of health, it can be estimated from cumulative throughput and cycle count if both are readable over the interface, which is a further argument for specifying the richer interface at purchase rather than retrofitting it.
Degradation: Where the 20 Percent Actually Goes
Operators routinely report that a fleet's packs diverge in health far more than the specification predicts. The divergence is usually traceable to four causes, in this order of frequency.
- Unequal route assignment. The robot on the longest or hilliest route accumulates throughput fastest. Over eighteen months this alone can produce a 15-point spread in state of health across nominally identical packs.
- Charge-window heat. A pack charged at high rate in a warm plant room, with no gap between discharge and charge, sits at elevated temperature for hours per day. High-temperature time at high state of charge is the single largest driver of calendar ageing, and the mitigation is scheduling — a short rest before charging, and charging away from the warmest part of the facility.
- Storage at full charge. Units taken out of service and stored at 100 percent for weeks will lose capacity faster than units cycled daily. The correct storage state is around 40–60 percent.
- Cell-group imbalance. A pack with one weak cell group will see the BMS terminate discharge early to protect that group, which reduces usable capacity and accelerates the imbalance. Detectable from cell-group voltage spread read over the RS485 interface; not detectable from pack-level voltage.
The maintenance implication is direct: the pack interface that lets a fleet read state of health and cell-group spread is the same interface that makes the divergence manageable, because it allows route rebalancing and early replacement of one pack rather than a whole fleet at once. The measurement framework for tracking that across a fleet is set out in service robot KPI benchmarks, and the wider recurring-cost view — what a battery replacement does to total cost of ownership — is in maintenance and total cost of ownership.
Sizing the Pack to the Shift, Not to the Datasheet
The most common specification error is sizing a pack to the working shift rather than to the working shift plus the charge windows available within it. The correct sequence runs as follows.
Run in that order, the pack specification becomes a consequence of the duty cycle rather than a number carried over from a previous project. Run in the opposite order — pick a pack, then discover the duty cycle does not fit — and the result is the familiar pattern of a fleet that performs to specification in the pilot and underperforms once the routes are real.
Sourcing: The Year-Four Question
Service robots are capital assets with a five-to-eight-year planning horizon, and a battery is a consumable inside that horizon. The sourcing question is therefore not only what the pack costs today, but whether a compatible pack can be bought in year four without redesigning the robot.
| Question | Good answer | Warning sign |
|---|---|---|
| Cell format and supplier | Commodity cylindrical cells with at least two qualified suppliers | Proprietary format from a single supplier with no second source documented |
| BMS protocol documentation | Published register map, versioned | Protocol available only under NDA, or not at all |
| Mechanical envelope | Documented dimensions, mounting points and connector part numbers | Pack is a sealed assembly with no serviceable connector |
| Capacity continuity | A stated minimum availability commitment, or a documented drop-in successor part | No commitment; the answer is that the pack will be quoted at the time |
| Certification | Transport and safety documentation held by the supplier, with certificates issued in the buyer's name where needed | Documentation is the buyer's responsibility to obtain |
The connector and envelope questions deserve particular attention because they are the ones that turn a routine consumable replacement into an engineering project. A pack whose dimensions and connector are documented can be re-sourced from any qualified supplier; a pack sealed into the chassis cannot, and the fleet owner's negotiating position at replacement time depends entirely on the alternative existing.
The unit economics of a replacement cycle are worth stating plainly. A service-robot pack typically represents a meaningful fraction of the platform's capital cost, and it will be replaced at least once in a normal service life. A buyer who specifies a re-sourceable pack with a documented interface is buying a replacement at market price in year four. A buyer who does not is buying whatever the original supplier quotes. That difference is larger than most of the per-unit hardware decisions that occupy more attention in a procurement process.
The Battery Specification in Five Lines
Reduced to the minimum a purchase order should carry, the specification is this: chemistry selected against the measured duty profile and ambient range, not against energy density alone; continuous and pulse discharge ratings sized from the peak load budget with margin at the worst-case ambient; charge acceptance curve provided, because it determines what a dock window actually returns; RS485 with a published register map reporting state of charge, state of health, cell-group voltages and cycle count; and a documented mechanical envelope and connector so the pack can be re-sourced independently in year four.
the charging-architecture companion to this article — dock placement, opportunistic charging and fleet power planning — is covered in service robot battery and charging technology and fleet charging infrastructure and power planning; for how a robot platform behaves when a pack is swapped out of service, see downtime cost and OEE.How AOMAN Approaches Battery Specification
AOMAN FUTURE builds the D1 delivery robot, the C1 large-format cleaning robot, the C2 Pro compact cleaner and the G1 reception robot in Shenzhen, and battery specification is handled as a per-deployment decision rather than a fixed catalogue item. The platform family uses LFP chemistry for the indoor cleaning and delivery duty cycles that dominate our volume, with pack-level specifications — continuous and pulse rating, charge acceptance, and an RS485 interface carrying state of charge, state of health and cell-group voltages — documented for each configuration so a fleet operator can read battery health into their own management layer rather than through a proprietary tool. Where a deployment runs cold storage, outdoor winter routes or three-shift opportunity charging, the chemistry recommendation changes and we will say so rather than quote the default.
The practical offer is narrower and more useful than a datasheet. Send us the duty cycle — shift length, route profile, ambient range, peak actuator load and the dock windows available — and we will return a pack specification derived from those numbers, with the mechanical envelope and connector part numbers documented so the pack can be re-sourced independently at replacement time. Send us the duty profile and we will run the sizing arithmetic against it. For the wider platform engineering context, see AOMAN technology and the Shenzhen production facility.
