Blog · 2026-07-05 · Power
Between the battery and everything else sits the least glamorous subsystem on the robot — and the one that decides whether the glamorous parts work. Here's the architecture that experienced builders converge on, and why each piece is there.
Nobody starts robotics to think about power distribution. You start for the motors, the arms, the autonomy — and then you spend a weekend chasing a "software bug" that turns out to be a servo dragging the 5 V rail down through a regulator that was never going to cope. Power distribution is the subsystem where an hour of design saves whole weekends of ghost-hunting, and the good news is that the design space has already been explored: hobby robots of every size converge on the same architecture. This article walks through it, part by part.
Every workable robot power system separates its loads into two families. The dirty rail is raw battery voltage feeding the hungry, noisy things: drive motors through their driver, big actuators, anything that spikes. The clean rail is regulated 5 V (or 3.3 V) feeding the sensitive things: microcontroller, sensors, radio. The two rails share exactly one thing — ground — because signals between subsystems need a common reference. Motors dump electrical noise and current spikes into their supply; keeping them off the logic rail means their tantrums stay in their own room.
Battery ─┬─ fuse ─ switch ─┬─ motor driver ─ motors (dirty)
│ └─ UBEC ─ 5 V ─ MCU, sensors, servos (clean)
└────────────── common ground everywhere ──────────────
Everything else in this article is detail on that diagram: which regulator, which fuse, which wire, in which order.
The regulated rail needs a regulator, and the hobby world's name for it is the BEC — battery eliminator circuit, a term inherited from RC aircraft where it eliminated the separate receiver battery. A UBEC is the same thing as a standalone module, and the U might as well stand for "use this one."
The distinction that matters is linear versus switching. A linear regulator (the classic 7805, the regulator on an Arduino) makes 5 V by burning off the difference as heat: feeding it 11.1 V at 1 A of load means dissipating 6.1 W — a soldering-iron amount of heat from a fingernail-sized part, which is why it thermally limits long before its printed rating. A switching regulator converts at 85–95% efficiency: the same job wastes under a watt, and 1 A on the 5 V rail costs only ~0.5 A from a 3S battery. For anything beyond a bare microcontroller, switching is the only serious option. Size it at 1.5–2× the clean rail's worst-case load — the Power Budget Calculator totals the load and picks the rating — and prefer name-brand 5 A/8 A/10 A UBEC modules with actual heat-shrink-covered inductors over mystery three-pin parts.
Servos confuse the two-rail model because they're noisy like motors but need regulated voltage like logic. The resolution: servos get regulated power from a dedicated or generously-sized BEC, with only their signal wires visiting the microcontroller. Standard servos run on 5–6 V (6 V gives measurably more torque and speed — check the datasheet); a robot arm's worth of them can draw 4–8 A in a coordinated move, which is a full-sized UBEC's job, not a shared one. High-voltage servos that accept 2S battery voltage directly simplify life further by moving themselves onto the dirty rail entirely.
A fuse protects the wire, not the electronics — by the time a fuse blows, your transistors have long since voted. Its job is preventing the catastrophic case: a hard short across the battery turning wiring into a heating element inside a plastic chassis. One automotive blade fuse in the battery positive lead, rated ~1.4× your continuous draw (so it never nuisance-blows on motor starts) and below the wire's failure current, is the minimum standard. Larger robots fuse each branch — drive, arm, logic — so one fault kills one subsystem instead of the match. Add a main switch rated for the full current (or an XT60 loop key, the competition-standard removable link), because "unplug the battery" is not an emergency stop when a robot is misbehaving in your hands.
Two smaller protections earn their pennies: a reverse-polarity guard (keyed connectors like XT60 mostly solve this; a P-FET circuit solves it completely) and a low-voltage alarm on the balance connector, because the power system's job includes protecting the battery from you.
With the parts chosen, their arrangement decides the noise behaviour. The principle is star grounding: pick one point — the battery negative, or a power distribution board's ground plane — and give every subsystem its own path back to it, rather than daisy-chaining grounds through each other. The reason is Ohm's law again: ground wires have resistance, and a motor's 10 A returning through a shared ground segment lifts that segment by tens or hundreds of millivolts — which the microcontroller, referencing the same "ground," reads as noise on every sensor. Daisy-chained grounds are how a motor PWM ends up visible in an analog sensor reading three connectors away.
The physical layout rules follow the same logic: keep high-current loops (battery → driver → motor → back) short and their wires paired or twisted; route signal wires away from motor wires, crossing at right angles when they must meet; put the UBEC's input close to the battery and its output close to the loads. And decouple locally — 100 nF ceramic capacitors at each IC's supply pins, plus a few hundred µF of bulk electrolytic where power enters each board — so brief demands are served from local storage instead of yanking the whole rail.
Concreteness helps. Take a 3 kg rover: two gearmotors (3 A each working, 12 A stall), a pan-tilt with two standard servos (~1 A each under load), an ESP32 with sensors (~0.4 A). Architecture: 3S LiPo → 15 A blade fuse → XT60 loop key → distribution board. Dirty rail: dual motor driver powered directly, driving the gearmotors on 14 AWG-to-driver, 18 AWG-to-motor wiring. Clean rail: 8 A switching UBEC making 5.5 V for the servos and 5 V bus; ESP32 fed from the bus through its own 3.3 V regulation. Total battery draw ≈ 6 A working (the calculator's math: motors 6 A + 2.4 A of 5 V load costing ~1.3 A at the battery ≈ 7.3 A — fuse and pack verified against it via the C-Rating Checker). Every number traceable, every part sized on paper before soldering. That's the whole discipline.
New power systems deserve a ritual. First power-up with no loads connected: verify polarity at every connector with a multimeter, verify the UBEC outputs 5.0-ish volts, feel for anything warming. Then add the microcontroller alone. Then sensors. Then servos, centered, one at a time. Motors last, wheels off the ground, then loaded. At each stage, a current meter inline (or a bench supply with a display, for the pre-battery stages) tells you if reality matches the budget. Smoke happens at step one or never; this ordering makes it step one, with nothing expensive attached.
Any robot with more than three or four power consumers benefits: a PDB (or even a screw-terminal bus bar pair) replaces the "solder blob of many wires" with inspectable, modifiable connections and often includes the star-ground point for free. Drone PDBs with integrated BECs are cheap and excellent for small robots.
Yes — that's the entire two-rail architecture. Separate batteries (one for logic, one for motors) is the old-school alternative that still has its place in noise-critical or beginner builds, at the cost of two packs to charge and monitor. Shared pack + good UBEC + star ground achieves the same peace more elegantly.
Servo torque and speed scale with voltage, and most standard servos are rated to 6 V (check!). Many UBECs have a 5 V/6 V jumper — the 6 V setting is a free ~15% torque upgrade for the servos while the logic keeps its own 5 V or 3.3 V supply.
Two rails, one ground, a switching UBEC, a fuse that protects the wire, and a bring-up ritual — that's the architecture. Size every element of yours in the Power Budget Calculator, then make the connections worthy of the design with the companion article, Wire Gauge and Connectors for Robots.