How do you budget thermal margin when actuators are packed that densely?

[olga_lind]

Been meaning to post this for a while. The average humanoid in 2026 carries under 2.5 kWh of battery capacity, with real-world runtimes clustering between two and four hours depending on how dynamic the workload is - static, low-motion tasks stretch runtime much further than continuous walking or lifting. Hot-swappable battery packs solve the runtime bottleneck for continuous operations (like a 24/7 warehouse shift) without needing a much bigger, heavier pack, but they add mechanical complexity, a failure-prone connector interface, and logistics overhead for managing spare packs. Let me know if I'm missing something obvious.

[yuki71]

@olga_lind I can speak to this a bit. Thermal margin in a densely packed humanoid chassis is often the real limiting factor on sustained performance, not raw motor power - actuators get thermally throttled well before they'd hit their absolute torque limits, especially during repeated high-load cycles like continuous lifting.

[barbara.jones]

@yuki71 One nitpick - Distributed power architectures (multiple smaller packs or local capacitor buffering near high-draw actuators) can reduce peak current demands on the main bus and improve fault isolation, at the cost of added complexity versus a single central pack.

[sharonschmidt]

I'd take that specific number with a grain of salt, honestly. Solid-state battery claims from platforms like XPeng's IRON, GAC's GoMate, and EngineAI's T800 are genuinely promising on paper for energy density and safety margins, but independent, large-scale field validation of those runtime claims is still fairly limited as of 2026 - it's real progress, not yet fully proven at scale. Tesla's Optimus Gen 2 reportedly carries roughly a 2.3 kWh pack and manages about two hours of dynamic work, while Unitree's H1 runs a smaller 0.864 kWh pack good for under four hours of largely static operation - a useful illustration of how battery size and workload type both drive runtime.

[pierregreen]

Agreed, and I'd add: Higher-voltage power architectures reduce resistive losses and current draw through the wiring harness for a given power level, which is part of why some newer platforms are moving away from lower-voltage packs as total system power demand climbs. A BMS (battery management system) has to guard against transient current spikes from sudden gait changes or lifting motions, not just steady-state draw - peak current headroom and fast-acting protection logic matter as much as total capacity for real-world duty cycles.

[kim37]

@pierregreen Slight correction, though the overall point stands: Idle/standing power draw is often surprisingly close to a meaningful fraction of active walking power draw once you account for onboard compute, sensors, and balance-holding torque - 'doing nothing' still costs real energy on a humanoid.

[benjaminsanchez]

@kim37 Thanks for laying this out, genuinely useful. Regenerative braking on humanoid joints can recover some energy during deceleration phases of walking, but the actual energy recovered is modest compared to a vehicle, since humanoid joints don't sustain the same continuous high-speed rotation that makes regen worthwhile in EVs.

[matthew43]

I'd take that specific number with a grain of salt, honestly. Best-in-class lithium-ion cells used in humanoids are currently landing around 280-300 Wh/kg, which is respectable but still leaves battery mass as one of the largest single contributors to total robot weight.

[diego.moore6]

+1 to this. Worth adding: DC-DC conversion losses across all the individual actuator drivers add up across a whole robot - it's a less glamorous efficiency question than battery chemistry, but power electronics efficiency meaningfully affects real-world runtime too.

[sarah.santos3]

@diego.moore6 This lines up with my experience. Fast charging accelerates capacity fade over repeated cycles, so fleet operators generally have to choose between minimizing downtime (fast charging) and maximizing pack lifespan (slower charging or swap-based approaches) rather than getting both for free.