Skip to main content

Battery Systems & Power Architecture

Duration: 45 min · Level: Intermediate · Module: 2. Actuator Architecture · Focus: battery, power, energy, systems

Your actuator map from the last lesson tells you how much torque the robot can produce. This lesson answers the question that actually decides whether the robot is useful: how long can it run before it has to stop and charge? G1 targets 8+ hours — a full hospital shift or home-care day — and hitting that number is an energy-budgeting problem you solve at the architecture level, not a battery you bolt on at the end.

Start from the power budget, not the battery​

The right order is: estimate average power, multiply by hours, then go shopping for cells. A 40-joint humanoid at moderate activity draws roughly 800 W to 1.5 kW average. For 8 hours that implies 6.4–12 kWh of usable capacity. That range is enormous, and closing it is the whole game — which is why duty-cycle optimization (most joints are near-idle most of the time) is as important as the cells you pick.

The mass problem, stated honestly​

Energy density sets a hard floor on weight. Lithium-ion 21700 cells deliver about 270 Wh/kg, so a naive 12 kWh pack weighs around 44 kg — more than the robot can carry and still be useful. The practical target is closer to 6 kWh, reached by driving the average power down with duty-cycle management rather than by adding cells. Two levers move that number: efficient actuators (Lesson 2.1's torque-density work) and software that lets idle joints rest.

There is also a chemistry choice with a safety dimension. Tesla Optimus uses lithium iron phosphate (LFP) cells — slightly lower energy density, but markedly safer in an impact or thermal event. For a robot working around people, that trade often favors LFP even at a small range cost. Decide this deliberately; it ripples into your thermal and safety design.

Architect the power delivery, not just storage​

How you move the energy matters as much as how you store it. The humanoid standard is a 48 V bus. Compared with 24 V, a higher voltage carries the same power at lower current, which cuts cable weight and slashes I²R losses in the wiring — a real efficiency win across dozens of joints. Pair that with modern GaN inverters, which reach 98%+ efficiency, and you keep more of every kWh doing useful work instead of heating wires.

Then reclaim energy you're already spending. Regenerative braking on the large joints — knee flexion and ankle dorsiflexion during stair descent — can recover 15–25% of locomotion energy by running the motors as generators. On a robot that walks all day, that is hours of extra runtime, not a rounding error.

Design for the shift change​

Even a perfect 8-hour pack eventually empties, so design the transition. A hot-swap architecture — two 3 kWh packs in parallel, one removable while the other keeps the robot alive — lets a caregiver change a battery in under two minutes without powering down. For a device meant to run continuous shifts, swap time is a product feature, not an afterthought.

Putting it into practice​

Size a real pack for your robot.

  1. From your Lesson 2.3 actuator table, estimate average power: assume each joint's continuous draw times its realistic duty fraction, and sum. This is your honest average, not the nameplate peak.
  2. Multiply by 8 hours to get required usable energy, then add ~20% for losses and reserve.
  3. Convert to mass at 270 Wh/kg (Li-ion) and again for LFP. Is the pack carryable? If not, identify which joints to duty-cycle harder.
  4. Choose a bus voltage and justify it; estimate the wiring-loss difference between 24 V and 48 V for your worst-case current.
  5. Sketch the hot-swap layout: two packs, the isolation/OR-ing scheme that keeps the robot live during a swap, and the target swap time.

Key takeaways​

  • Budget power first: a 40-joint humanoid averages ~0.8–1.5 kW, so 8 hours needs ~6.4–12 kWh — close that gap with duty-cycle optimization, not just bigger packs.
  • Energy density rules mass: 21700 Li-ion at ~270 Wh/kg makes a 12 kWh pack ~44 kg; target ~6 kWh of usable energy instead.
  • Chemistry is a safety choice: LFP (as in Tesla Optimus) trades a little density for much better impact/thermal safety — often worth it near people.
  • Architect delivery: a 48 V bus plus GaN inverters (98%+) cuts losses; regenerative braking recovers 15–25% of locomotion energy.
  • Design the shift change: hot-swap dual packs (e.g., 2×3 kWh) for sub-2-minute battery changes without downtime.

← Previous: 2.3 Joint-Level Architecture for a Full Humanoid · Next: 2.5 Thermal Management & Actuator Reliability →

Part of Module 2: Actuator Architecture.