Cases Display: 3.7V Lithium Battery for Robots

A 3.7V lithium cell is the smallest building block in any modern robot battery. It is also the most misunderstood. Hardware engineers spec it for mobile robots, AGVs, and medical carts expecting the cell’s nominal voltage to define what the pack can do — and then discover, often during integration testing, that the pack trips the BMS on every motor-startup cycle or sags below the controller’s cutoff voltage halfway through a shift.

Here’s the thing: the difference between a working 3.7V robot pack and a dead-on-arrival one almost never comes down to the cell itself. It comes down to whether the engineer sized the pack around continuous discharge current (C-rate), voltage sag under load, multi-cell topology, and the BMS communication protocol the robot’s controller actually expects.

This guide walks through that decision chain — from “why a single 3.7V cell is rarely enough” through multi-cell sizing, BMS integration, temperature derating, and two real application cases (one AGV, one portable medical device). For teams further down the sourcing path, our existing breakdown on How to Order Custom 3.7V Lithium Packs for Industrial Use covers the procurement and supplier side.

When a Single 3.7V Cell Is Enough (and When It Isn’t)

The nominal voltage of a single lithium-ion cell — 3.6V or 3.7V depending on chemistry, with a 4.2V full-charge ceiling and a 2.5–3.0V cutoff — is well-defined. What is not well-defined is whether a single cell can power the robot you’re actually building.

A single 3.7V cell (1S configuration) is the right call in a narrow band of robot applications:

• Sensor nodes and edge AI inference boxes drawing 1–10W continuous. A single 18650 at 3,000mAh gives you 11.1Wh — enough for days of operation at low duty cycle.

• Tiny inspection crawlers under 5kg total weight, with peak loads under 20W.

• Wearable telepresence or assistive robots where the enclosure physically cannot accept a multi-cell pack.

A single 3.7V cell is the wrong call the moment your robot:

• Drives traction motors that pull more than 5A peak at startup.

• Runs an onboard computer, LIDAR, and motor controller simultaneously.

• Needs a motor-controller input voltage above 5V (most motor controllers brown out below 5V under load).

If your robot falls in the second category — which is most industrial and mobile robots — you need a multi-cell configuration. The question is which one.

Robot Power Classes: What the Cells Actually Have to Deliver

Before you size a pack, you have to know what your robot asks for. Here’s a baseline table we use when engineers come to us with a “what do I need” question. Numbers are typical for real products, not theoretical extremes.

The numbers above come from real DNK Power engagements and from published specs on tier-1 industrial robot platforms. Two important notes:

1. Peak power is what kills undersized packs. A mid-size AGV rated at 800W continuous may pull 2,500W during acceleration. If you sized the pack for 1C continuous based on the 800W figure, your BMS will trip at every motor-start event.

2. Small mobile robots look like they could run on 3.7V — until you count peripherals. A service robot with a 50W drive motor, 25W onboard computer, 10W LIDAR, and 5W comms radios needs ~90W continuous. At 3.7V nominal, that’s 24A continuous from a single 18650 — outside any commercial 18650’s safe operating area.

For a deeper pass at the broader 12V robot platform — which is the most common migration target from single 3.7V cells — our 12V battery guide for robots covers the chemistry and C-rate trade-offs in detail.

Why a Single 3.7V Cell Has a Power Ceiling

Let’s do the math. A quality 18650 cell (Samsung INR18650-35E, Panasonic NCR18650B, or equivalent) delivers 3,000–3,500mAh at 3.6–3.7V nominal. That’s roughly 11–13Wh per cell. The continuous discharge rating of a high-drain 18650 is typically 8–10A, with a peak of 15–20A for short bursts (≤5 seconds).

Translated to power:

• Continuous: 3.7V × 10A = 37W per cell, at the upper edge of the cell’s thermal envelope.

• Peak (≤5 sec): 3.7V × 20A = 74W per cell, before voltage sag and derating kick in.

The continuous discharge ratings cited above are typical for high-drain 18650 cells; actual ratings vary by manufacturer and lot, and the 18650 form factor spans chemistries (LCO, LMO, NMC, LFP) with very different voltage and discharge profiles. For a baseline reference on the form factor, see the 18650 battery entry on Wikipedia or the manufacturer datasheet for the specific cell in use.

Two real-world consequences:

1. Voltage sag under load. A cell pulling 15A sag is closer to 3.2V at the terminal — a 14% drop from nominal. Your motor controller, expecting 3.7V, may brown out.

2. Thermal derating. Pulling 10A continuous from a single 18650 heats the cell 15–20°C above ambient. Above 45°C cell-surface temperature, the cell’s effective capacity drops by 15–25% and the cycle life starts shortening.

So the upper practical power limit for a single 18650 cell, with reasonable thermal margin and acceptable cycle life, is roughly 30–40W continuous. Anything beyond that needs parallel cells, higher voltage (series), or both.

Multi-Cell Configurations: Series, Parallel, or Both

Once you’ve decided a single cell is not enough, the next decision is the topology. Three options exist, each with a different electrical outcome.

The key insight: series multiplies voltage, parallel multiplies capacity. Most robot platforms need both — voltage high enough to drive the motor controller efficiently, and capacity high enough to last a shift.

A practical example: an AGV traction motor rated for 24V nominal and pulling 30A continuous needs 720W of continuous power. At 25.9V (7S) and 30A, the pack needs to deliver 30A continuously — which means a 7S configuration with cells rated for at least 1C continuous per parallel string. To hit 8 hours of runtime at 720W, you need roughly 5,760Wh of usable capacity, factoring in 80% depth-of-discharge — a 7S20P or 7S30P pack at the cell level, depending on the cell you choose.

For most AGV and AMR programs, a 7S to 14S NMC or LiFePO4 topology is the practical sweet spot. Above 14S, you cross into 48V+ systems, which require additional touch-safe considerations and different BMS hardware.

BMS Integration: The Protocol Question Robot Engineers Forget

A lithium battery pack without a BMS that the robot’s host controller can read is a black box. The robot doesn’t know state-of-charge, doesn’t know state-of-health, and doesn’t see faults until the pack dies mid-shift.

For a robot battery pack, the BMS has to do four things:

1. Cell protection. Overvoltage and undervoltage cutoffs depend on cell chemistry and manufacturer — typical NMC thresholds are 4.20V/cell overvoltage and 2.80V/cell undervoltage, but always verify against the selected cell’s datasheet. Overcurrent cutoff is commonly set around 110% of the continuous load, with short-circuit response under 500µs as a baseline target for industrial packs.

2. Cell balancing. ±50mV tolerance at full charge is a common target range for most AGV and service-robot packs; tighter tolerances (e.g., ±20mV) require active balancing hardware. Passive balancing is sufficient for the duty cycle of most mobile robots; high-cell-count packs in long-runtime applications benefit more from active balancing.

3. Thermal management. Cutoff at 60°C for most packs, with low-temperature charge lockout at 0°C for NMC and –10°C for LiFePO4 to prevent lithium plating.

4. Communication. Expose state-of-charge, state-of-health, cell voltages, temperatures, and fault codes to the host controller over a recognized industrial protocol.

That fourth item is where most projects stall. The two protocols that dominate robotics and AGV integration are CAN Bus (CANopen / J1939) and RS485 Modbus RTU. Here’s how they compare for robot battery applications:

For a robot with a CAN-enabled motor controller — which is most modern AGV and AMR platforms — CANopen with a CiA 418 battery profile is the lowest-friction integration path. For a robot with a low-cost MCU on a tight firmware budget, RS485 Modbus is often the practical choice, but expect to define and validate the register map on the supplier side.

For the full decision framework on protocol selection, message-map design, and host-side validation, our BMS and inverter communication integration case study walks through a real engagement that covered both protocols.

Temperature Performance: Why Cold Warehouses and Hot Foundries Change the Math

A 3.7V lithium cell rated for 25°C continuous operation does not deliver the same capacity at –10°C or +50°C. Engineers who size the pack from a datasheet spec sheet without applying a temperature derating factor discover this on the first cold morning or summer heat wave.

Typical derating for an NMC 18650 cell at moderate discharge rates (≤1C):

The percentages above are representative values for NMC 18650 and 21700 cells at moderate discharge rates (≤1C). Actual capacity retention at temperature extremes varies by cell manufacturer, cell format, age, and discharge profile. Always confirm the derating factor against the cell datasheet for the specific lot in use.

Two practical consequences:

1. Cold-storage AGVs (–20°C to –10°C) need a heating pad or thermal management system. A passive pack without heating will deliver 60–70% of rated capacity and may refuse to charge entirely if the cell temperature is below 0°C. Lithium plating on the anode — caused by charging below 0°C — is a permanent damage mechanism, not a recoverable fault.

2. Outdoor or foundry-adjacent robots (above 45°C) need derated capacity targets. Sizing for 100% rated capacity at 50°C means 85% real-world capacity. Build the buffer into the spec, not into the post-delivery field report.

The same BMS that handles protection also handles thermal cutoffs. A well-designed industrial BMS locks out charging below 0°C (NMC) or –10°C (LiFePO4) and reduces max discharge current above 55°C to keep the cells inside their safe operating area.

Case 1: Mid-Size AGV on a Custom 24V NMC Pack

A mid-size AGV OEM came to DNK Power with a duty cycle the catalog pack could not meet. The platform was a 200kg payload mobile base used in a parts warehouse, running two 8-hour shifts per day, with 30-minute fast-charge windows between shifts.

The original spec called for a “24V 60Ah” pack. The supplier’s catalog 24V LiFePO4 pack delivered the right capacity, but the BMS topped out at 1C continuous discharge — not enough for the AGV’s 800W traction load, which pulled 33A at 24V nominal and peaked at 90A during acceleration. The motor controller would brown out on every lift cycle.

DNK Power’s engineering team scoped a custom 7S20P NMC pack using Molicel P42A cells (4.2Ah per cell, rated 30A continuous discharge per cell — roughly 7C for that specific cell). At the pack level, that translated to 600A continuous capability (about 15.5kW at the pack’s 25.9V nominal) and 900A peak for 5 seconds — well beyond the AGV’s 800W continuous / 2,500W peak demand. The BMS was configured to a 300A continuous limit and a 500A peak window (giving roughly 4× headroom over the worst-case acceleration surge), exposing CAN Bus with a CANopen CiA 418 message map that the OEM’s motor controller already supported.

Results on the integration test bench:

• Continuous discharge capability: 300A (vs the catalog pack’s 60A)

• 0–80% state-of-charge recharge in 28 minutes at the AGV’s 2C charge station

• Cold-warehouse operation validated at 5°C ambient with no heating pad (cells kept above 10°C by the AGV’s waste motor heat)

• 18-month field deployment with zero BMS-trip events across 24 units

The OEM’s biggest lesson was that “24V 60Ah” was a capacity spec without a discharge-rate spec — and the catalog pack was inadequate for the actual duty cycle the AGV imposed.

Case 2: Portable Diagnostic Cart on a Custom 11.1V LiPo Pack

A medical device manufacturer developing a portable diagnostic cart — basically a wheeled cart with an integrated analyzer, touchscreen, and battery — needed a slim pack that fit a 25mm-thick enclosure bay. Cylindrical 18650 cells were mechanically out. LiPo pouch cells were the only option.

The spec called for 11.1V nominal (3S LiPo), 5,000mAh minimum, 3C continuous discharge for the analyzer’s internal pump, and a flat rectangular footprint that no off-the-shelf pack matched.

DNK Power built a custom 3S2P LiPo pack using industrial-grade 5,000mAh pouch cells rated for 5C continuous. The BMS was configured for RS485 Modbus RTU to match the analyzer’s existing host controller — no firmware rewrite needed on the medical side. The pack was UN38.3 certified at the cell level and shipped with the IEC 62133 cell-level report the FDA submission pathway required.

Three practical takeaways from this engagement:

1. LiPo pouch cells in industrial packs require compression frames. Without a rigid mechanical frame around the cells, swelling over 200–400 cycles can deform the pack and stress the welds. The compression frame adds 1–2mm to the pack thickness and is non-negotiable for any product with a multi-year field life.

2. Pouch cell BMS thresholds often need to be tighter than cylindrical cell thresholds. Pouch cells are typically less forgiving of overdischarge (below 3.0V/cell) and overcharge (above 4.25V/cell) than cylindrical cells with the same chemistry, though the exact limits vary by manufacturer. Set BMS voltage-tolerance windows against the specific cell datasheet rather than a generic LiPo spec.

3. Industrial-grade LiPo (typically 5C continuous, in gel-polymer or laminated industrial formats) costs more than hobby-grade LiPo (often rated 25C continuous), but typically delivers 3–5x the cycle life. Hobby cells commonly rate 300–500 cycles at 80% depth-of-discharge; industrial cells often rate 800–1,200 cycles at the same depth-of-discharge. The exact multiplier depends on the manufacturer, the depth-of-discharge profile, and the operating temperature — confirm against the cell datasheet before sizing.

For the full procurement and supplier-evaluation playbook on these kinds of custom 3.7V-class packs, our case study on medical and AGV OEM pack programs covers the supplier-selection criteria we used.

The DNK Power 24-Hour Custom Design Flow

For teams whose robot spec doesn’t match a catalog pack — which is most of the AGV and medical-cart programs that come to us — the path from spec to functional sample runs through a fixed set of steps:

1. Spec submission. Nominal voltage, capacity, continuous and peak discharge, physical envelope, connector, BMS protocol, operating temperature, target certifications.

2. 24-hour design package. Within one business day, the engineering team returns a mechanical fit-check, electrical spec alignment, BMS configuration scope, and a fixed quote.

3. 3-day functional sample. For most custom pack programs, the first working sample ships in three business days from spec lock, with prototype BMS firmware loaded.

4. Validation. Protocol conformance test against a reference host simulator, fault injection test on all declared protection thresholds, and a thermal characterization at the customer’s worst-case ambient.

5. Certification scoping. UN38.3 test summary, IEC 62133 cell-level report, MSDS, and pack-level UL 2054 or CE technical file, scoped against the customer’s target market.

6. Pilot run. 50–500 units, with the same lot traceability and firmware version as the mass-production run.

7. Mass production. With a 100% refund guarantee if the production run drifts from the approved sample.

The compression of the design-and-sample step from the industry-standard 4–6 weeks to 24 hours for the design package and 3 days for the first functional sample is the single biggest lever on launch timeline for new robot platforms. It is not a marketing line — it is the working rhythm our engineering team has built around a fixed set of design review steps.

Pre-Build Checklist: What to Lock Down Before Spec Submission

Before you send a spec to any custom battery assembler, lock down these inputs:

• Voltage window. Confirm the motor controller’s input voltage range, including low-battery cutoff behavior. A 24V controller that browns out below 22V needs the pack sized so nominal voltage stays above 24V through 80% of the discharge cycle.

• Continuous and peak current. Worst-case inrush (motor start, actuator fire) divided by capacity to get the required C-rate. Don’t size for continuous and hope peak is fine — it won’t be.

• Capacity target in Wh, not Ah. Wh is voltage-agnostic. A “24V 60Ah” spec is meaningless without a C-rate; “1,440Wh at 5C continuous” is actionable.

• Operating temperature range. Indoor (15–35°C) is easy. Outdoor, cold storage, or foundry-adjacent (–20°C to 50°C) requires thermal management.

• Form factor. Max dimensions in mm, mounting points, connector placement, wiring routing. Sketches help.

• Communication protocol. CAN Bus, RS485, RS232, SMBus — usually forced by the host MCU.

• Target certifications. UN38.3 for transport, IEC 62133 for safety, UL/CE/PSE/KC depending on geography.

If you can fill in those seven blocks with real numbers, you have a spec a manufacturer can quote against. If you can’t, you’re buying blind.

When to Stop Reading and Start Talking to a Custom Assembler

A 3.7V single-cell pack is the right answer for a narrow band of robot platforms — sensor nodes, edge AI inference, ultra-light wearables. For everything else, the question is not “should I go custom?” but “how fast can my supplier turn the spec into a validated sample?”

If you have a robot platform that needs a custom 3.7V-class pack — single cell, multi-cell series, multi-cell parallel, or a series-parallel hybrid — and you want a design package back in 24 hours and a functional sample in three days, send the spec to DNK Power’s engineering team. The faster you lock down the seven inputs above, the faster the first sample lands on your integration bench.