{"id":10842,"date":"2026-11-06T08:32:30","date_gmt":"2026-11-06T08:32:30","guid":{"rendered":"https:\/\/www.herewinpower.com\/blog\/humanoid-robot-battery-selection-energy-density-power-weight\/"},"modified":"2026-11-06T08:32:30","modified_gmt":"2026-11-06T08:32:30","slug":"humanoid-robot-battery-selection-energy-density-power-weight","status":"publish","type":"post","link":"https:\/\/www.herewinpower.com\/ru\/blog\/humanoid-robot-battery-selection-energy-density-power-weight\/","title":{"rendered":"How to Choose a Battery for Humanoid Robots: Energy Density, Power and Weight"},"content":{"rendered":"
\"Illustration<\/figure>\n\n\n\n

A high-energy cell can meet its datasheet specifications and still prove unsuitable for a humanoid robot. The issue may not be insufficient stored energy, but the pack’s ability to maintain bus voltage during short, high-current movements.<\/p>\n\n\n\n

Consider a hypothetical case: a battery pack is sized to support three hours of average operation using a high-energy cell rated at 300 Wh\/kg. During a loaded squat-to-stand movement, however, the current demand exceeds the pack’s transient capability, causing the DC bus voltage to fall below the drive’s undervoltage threshold. The robot stops even though substantial energy remains in the battery.<\/p>\n\n\n\n

The example illustrates a common sizing problem: average energy consumption and short-duration power demand must be evaluated separately. Wh\/kg alone cannot predict this behavior, because voltage sag depends on cell construction, discharge rate, temperature, state of charge and the resistance of the complete pack.<\/p>\n\n\n\n

Humanoid robots combine continuous electrical loads with cyclic movement and brief high-power actions. Compute, sensors and control electronics draw power throughout operation; walking and posture control add changing loads; and movements such as standing up, jumping or recovering balance can produce much higher current peaks. A cell tested only at 0.5C provides limited evidence of how it will perform under these conditions.<\/p>\n\n\n\n

Battery selection therefore needs to consider energy density, power capability, DCIR, peak current and pack mass together. Each of these factors, and the validation work an OEM needs to carry out on the assembled pack, is covered below.<\/p>\n\n\n\n

Understand the Power Demands of Humanoid Robots<\/h2>\n\n\n\n

The measurements cited in this section come from Fraunhofer IPA’s whole-robot power measurement of a Unitree G1 EDU-4<\/strong><\/a> with Dex3-1 hands and firmware 1.04. They are mean electrical power over each scenario, not instantaneous peaks, and the study publishes no peak-power figure for this platform.<\/p>\n\n\n\n

Continuous Loads and Baseline Power Consumption<\/h3>\n\n\n\n

A humanoid robot consumes power before it moves. Compute, perception sensors, control electronics, communication and thermal management draw continuously. In many humanoid platforms, the hip and knee actuators must also continuously produce holding torque to maintain a standing posture, particularly where the joint design does not provide mechanical self-locking or a brake that can carry the load. Platforms that use different transmissions, brakes or mechanical holding arrangements will show a different standing baseline.<\/p>\n\n\n\n

In stationary operation, that G1 configuration measured approximately 154 W average consumption. The figure covers the whole robot at rest, not one subsystem, and it is a measurement for that configuration under Fraunhofer’s test conditions \u2014 not a constant for the category.<\/p>\n\n\n\n

A second baseline effect matters for sizing. Extending the G1’s arms horizontally without added payload caused the actuators to overheat and drop after roughly one to two minutes in the tested configuration. Holding a posture can therefore be a thermal constraint before it is an energy constraint: the joint draws sustained current while producing zero mechanical output, and the electrical energy becomes heat in windings and power electronics.<\/p>\n\n\n\n

Cyclic Loads During Walking and Posture Control<\/h3>\n\n\n\n

Walking superimposes a periodic load on the baseline. Fraunhofer IPA measured approximately 272 W while walking on level ground and around 283 W on a 10 percent incline, against the 154 W standing figure. For its defined one-hour standard scenario, the resulting average was approximately 239 W.<\/p>\n\n\n\n

The standing condition consumed about 57 percent of level-walking power in this benchmark. Treating the difference between standing and walking as “the power of walking” misstates the load. The periodic layer rides on a baseline that never switches off, so both have to be budgeted together.<\/p>\n\n\n\n

Walking power is also not smooth. Within each gait cycle, the hip, knee and ankle joints deliver torque pulses at push-off and absorb energy at contact. The average walking figure is an integration over those pulses. Current demand can rise sharply during push-off even when average walking power remains relatively modest, which is why a pack sized only against the walking average can still trip during ordinary locomotion.<\/p>\n\n\n\n

Short-Duration Peaks During Dynamic Movements<\/h3>\n\n\n\n

Short-duration power peaks are often the most difficult part of battery sizing. Rising from a deep squat, jumping, running, carrying a heavy payload or recovering balance after an external disturbance concentrates energy demand into hundreds of milliseconds.<\/p>\n\n\n\n

Published power budgets show the scale. A Monolithic Power Systems humanoid power architecture reference<\/a> lists hip joints at 600\u2013800 W average and 1500\u20132000 W peak, and describes leg subsystems drawing 3\u20134 kW per leg during burst conditions at a 48 V bus. A knee joint study on a 1-DoF leg platform reports that during a box jump the peak joint torque reached 286 Nm and peak joint mechanical power reached 1.5 kW \u2014 near the limits of the TQ-ILM8526SP actuator, whose motor-side peak torque is 9.37 Nm and nominal peak power is approximately 1.5 kW. The authors report short-term overload capability in the single-joint platform tests, but also observed loss of control beyond that range.<\/p>\n\n\n\n

These figures come from different platforms and test conditions, so they should not be combined into a single robot specification. They nevertheless show how sharply power demand can rise during dynamic movements, and why transient capability is a separate requirement: a battery that meets the baseline and average walking demand may still fall short if its terminal voltage drops below the drive’s operating threshold.<\/p>\n\n\n\n

Rated current does not define that requirement either. During landing, jumping or sudden acceleration, joint motors can draw well above their rated current for a short period. The actual figure depends on the actuator, the motor controller and its current limits, the bus voltage and the pulse duration, so it has to be established per robot rather than inferred from a motor rating or an average load.<\/p>\n\n\n\n

Balance Energy Density, Power Capability and DCIR<\/h2>\n\n\n\n

Wh\/kg and W\/kg: Two Different Performance Requirements<\/h3>\n\n\n\n

Gravimetric energy density, Wh\/kg, states how much energy a cell stores per unit of its own mass. Gravimetric power density, W\/kg, states how much power it can deliver per unit mass. Both figures are normally quoted at the cell level. They answer different questions: cell Wh\/kg sets the ceiling on runtime for a given battery mass, and cell W\/kg indicates the rate at which that cell can be discharged under defined conditions. Neither one, on its own, describes the finished pack.<\/p>\n\n\n\n

The two are not independent, and the coupling runs the wrong way for a humanoid designer. Increasing cell-level energy density often involves raising active-material loading, improving electrode utilization, or reducing the relative mass of inactive components. However, higher electrode loading and thicker electrodes can lengthen the diffusion path lithium ions must travel through the electrolyte and the solid particles, which raises polarization and internal resistance at high rate. A cell optimized for higher energy density may therefore offer less power capability under demanding discharge conditions. A cell built for the opposite trade \u2014 thin electrodes, high-rate current collection \u2014 gives up energy density to gain power.<\/p>\n\n\n\n

Neither figure transfers directly to the assembled pack. Cell-level Wh\/kg drops once you add enclosure, interconnects, BMS, harness, connectors and protective structure. Power degrades differently: the pack has to deliver current at the bus where the drives connect, through contactors, fuses and the BMS current path, without crossing the permitted voltage sag. A cell’s W\/kg cannot prove the pack meets a drive’s peak current demand.<\/p>\n\n\n\n

So a high-Wh\/kg cell and a high-W\/kg cell may both be suitable, depending on the robot’s motion profile. Screen cells on cell-level energy and rate data, then write the actual requirement at the pack level.<\/p>\n\n\n\n

How Cell Design Affects High-Rate Discharge<\/h3>\n\n\n\n

Several cell-design factors influence the balance between energy density and high-rate performance, regardless of cathode chemistry.<\/p>\n\n\n\n