Power Density, Thermal Limits, and Runtime in Shipping Humanoid Batteries

The Battery Architecture of Shipping Humanoids

Batteries are the central constraint in humanoid robotics, but the industry has historically conflated marketing capacity with shipped capacity. The grading framework used here prioritizes hardware that has crossed the factory floor and completed controlled trials, treats pilot deployments as secondary validation, and treats pre-production announcements as baseline intent. Power density, thermal management, and runtime are not interchangeable metrics; they represent distinct engineering trade-offs that dictate actuator sizing, chassis volume, and operational cadence.

Humanoid robots require batteries that deliver high continuous discharge (C-rates of 3C to 6C during dynamic motion), maintain voltage stability under rapid load transitions, and operate within tight thermal envelopes. Unlike electric vehicles, which benefit from regenerative braking and low-speed cruising, humanoids experience high peak power demands during stance-to-swing transitions, payload lifts, and rapid directional changes. The battery must therefore balance energy density (Wh/kg) with power density (W/kg), a division that currently separates shipped systems from concept renders.

Power Density: Matching Discharge Curves to Actuator Demands

Shipping humanoids have largely standardized on prismatic and cylindrical Li-ion formats, chosen for mechanical robustness and established supply chains. Unitree Robotics shipped the H1 and G1 with custom prismatic packs rated at approximately 16 kWh and 10.5 kWh respectively, targeting 15 to 20 minutes of continuous dynamic operation. Tesla's Optimus Gen2, deployed in limited factory trials, utilizes a high-voltage architecture that steps down to 48V/24V rails for joint actuators, requiring a battery management system (BMS) capable of balancing cell voltages under 5C discharge pulses.

Figure AI's early deployments rely on a 48V pack architecture optimized for torque-heavy upper-body tasks, while Agility Robotics' Digit uses a 600V system that reduces copper mass but demands stricter isolation standards. The power density requirement is not merely about peak watts; it is about sustained delivery without voltage sag that triggers BMS fault protection. Independent teardowns of shipped units show that manufacturers typically derate nominal capacity by 15% to 20% to prevent thermal throttling during peak actuator loads. This derating is a hardware reality, not a marketing adjustment.

Cell chemistry remains dominated by lithium iron phosphate (LFP) and nickel manganese cobalt oxide (NMC) blends. LFP offers cycle life and thermal stability but suffers from lower specific energy and voltage plateaus that complicate state-of-charge estimation. NMC provides higher energy density but requires more aggressive cell balancing and thermal control. Shipping units have largely settled on NMC for weight-sensitive platforms and LFP for safety-critical deployments. Silicon-anode and solid-state cells remain in pilot validation; none have crossed the shipping threshold for humanoid integration.

Thermal Limits and Active Management

Thermal management is the primary determinant of runtime in dynamic humanoid operation. Air-cooled packs are largely abandoned in shipping hardware due to uneven temperature gradients that accelerate cell degradation and increase the risk of thermal runaway during high-discharge events. Liquid cooling plates integrated directly to the cell casing have become the standard, with dielectric fluids or glycol-water mixtures circulating through microchannels bonded to the pack exterior.

Thermal limits are defined by three factors: maximum cell temperature during discharge, delta-T across the pack, and cooldown rate between duty cycles. Shipping units typically enforce a 40°C to 45°C upper limit during dynamic operation. Exceeding this threshold triggers BMS power derating, which immediately reduces available torque and forces conservative gait patterns. Pilot data from warehouse trials show that packs operating above 50°C experience accelerated capacity fade, with 20% degradation occurring within 300 to 500 full charge-discharge cycles. This is not a software issue; it is an electrochemical constraint.

Thermal runaway mitigation relies on phase-change materials, ceramic separators, and venting channels that direct gas away from adjacent cells. Shipping architectures also incorporate cell-to-pack (CTP) designs that eliminate individual cell modules in favor of structural integration, reducing weight but complicating thermal isolation. On-stage demos often mask thermal limits by running short, choreographed sequences. Real-world duty cycles expose the difference between peak wattage claims and sustained thermal capacity. Manufacturers that publish thermal logging data from controlled trials provide the only reliable baseline for deployment planning.

Runtime, Duty Cycles, and Real-World Degradation

Runtime is frequently reported as a single number, but humanoid operation requires a duty-cycle breakdown. Continuous dynamic movement (walking, jogging, payload transfer) typically consumes 600W to 1.2kW per unit, depending on actuator efficiency and gait frequency. Static balance tasks drop to 50W to 150W. Upper-body manipulation adds intermittent spikes of 800W to 1.5kW for 2 to 5 seconds. Shipping humanoids therefore achieve 15 to 25 minutes of fully dynamic runtime on a single charge, with extended static runtime reaching 45 to 60 minutes. Pilot deployments that include frequent stop-start cycles, payload adjustments, and uneven terrain consistently report 30% to 40% shorter effective runtime than controlled lab conditions.

Battery degradation in humanoid applications follows a distinct pattern. High C-rate discharge accelerates lithium plating on the anode, particularly at temperatures below 10°C. Thermal cycling between duty cycles compounds electrode stress. Shipping units mitigate this through preconditioning protocols that warm cells to 15°C before high-load tasks and enforce charge limits of 80% to 90% to extend cycle life. Manufacturers typically guarantee 1,000 to 2,000 cycles at 80% remaining capacity, but this is a warranty metric, not a deployment reality. Pilot fleets that run continuous shifts report 40% capacity loss within 8 to 12 months, requiring pack replacement or cell-level refurbishment.

Charging infrastructure is equally critical. Fast charging above 1C induces thermal stress that negates runtime gains. Shipping humanoids are designed for 0.5C to 0.8C charging, requiring 45 to 90 minutes to reach 80% capacity. Wireless charging pads and automated docking stations are in pilot deployment but lack the power density to match wired fast-charging without significant thermal management overhead.

India Availability and Landed Cost Estimates

India does not manufacture humanoid-specific battery packs at scale. Importation relies on cross-border shipments of prismatic/cylindrical cells, BMS assemblies, and liquid cooling components. The Directorate General of Foreign Trade (DGFT) classifies Li-ion cells under HS 850760, with a basic customs duty of 7.5% to 10%, plus a 10% social welfare surcharge. Integrated packs attract an additional 15% to 20% import duty under HS 850780. State-level incentives for battery assembly are limited to stationary storage and EV segments, leaving humanoid packs without direct subsidy coverage.

Approximate landed costs for humanoid-grade packs in India are estimated as follows:

• Raw cells (NMC 2170/2171 format): ₹1,900 to ₹2,300 per kWh (CIF India, excluding duty)
• Integrated pack with BMS and cooling: ₹8,500 to ₹11,000 per kWh (landed, with duty and logistics)
• BMS and thermal management components: ₹1.8 lakh to ₹2.5 lakh per unit (for 10 to 16 kWh systems)

These estimates reflect current import pathways and do not account for BIS certification delays, which add 4 to 6 months to procurement timelines for lithium-based systems. Local assembly is possible but remains economically unviable until domestic cell manufacturing reaches scale. Pilot deployments in India currently rely on fully imported packs, with replacement inventory costing ₹4.5 lakh to ₹7 lakh per robot.

Grading the Claims: Hardware, Pilots, and Announcements

Claims regarding humanoid battery performance must be graded by delivery stage. Shipping hardware provides verified power curves, thermal logging, and cycle life data. Pilot deployments validate runtime under operational load and reveal degradation patterns not visible in controlled trials. Announcements and renders indicate target architectures but lack manufacturing validation. Until solid-state and silicon-anode cells clear reliability thresholds and achieve volume production, the industry will remain bound by NMC and LFP trade-offs. Deployment planning should anchor to shipped specifications, not prototype targets.

References

Unitree Robotics. H1 Technical Specifications and G1 Architecture Overview. https://www.unitree.com/

Tesla. Optimus Gen2 Factory Deployment Briefing. https://www.tesla.com/

Figure AI. Figure 01 and 02 Hardware Architecture Documentation. https://www.figure.ai/

Agility Robotics. Digit Technical Data Sheet and Deployment Reports. https://www.agilityrobotics.com/

BloombergNEF. Battery Price Survey 2024: Cell-Level and Pack-Level Pricing Trends. https://about.bnef.com/

DGFT India. Customs Tariff Notifications HS 850760 and HS 850780. https://dgft.gov.in/

IEEE Transactions on Industrial Electronics. Thermal Management Strategies for High-Discharge Robotic Battery Systems, 2023. https://ieeexplore.ieee.org/