Technology deep-dive: humanoid-robot batteries and on-board energy storage
Access date for all URLs in this note: 2025-04-29.
Source-tier convention used here:
- Tier 1: granted patents, peer-reviewed papers, standards-body documents, regulator technical assessments.
- Tier 2: company technical docs/engineering blogs, named conference proceedings, independent benchmarks.
- Prohibited: SEO content farms, anonymous benchmark sites, AI-generated summaries, unverifiable secondary summaries.
Where a number could not be independently verified from an accessible primary source, it is written exactly as needs primary source validation.
Why onboard energy storage is the binding constraint for humanoid robots
A humanoid robot is an untethered, high-degree-of-freedom, power-dense machine. Its energy storage must serve two conflicting masters: energy density for runtime and power density for high transient loads during walking, stair climbing, arm payload moves, and recovery from a fall.
The energy-storage problem is not identical to electric vehicles, even though the underlying battery chemistry is usually lithium-ion. A humanoid pack is smaller than an EV pack, is cycled more often relative to capacity, is carried on a moving body where mass and thermal management matter, and may need hot-swapping or fast charging to keep a robot cell/fulfillment line operating 24/7. These requirements are technical claims, so they need source-bound evidence.
The oldest authoritative battery review used here, Dunn, Kamath, and Tarascon in Science, notes that battery choice is a trade space among energy density, power density, cycle life, safety, and cost (Science 2011, DOI:10.1126/science.1212741, published 2011-11-18, accessed 2025-04-29). Whittingham’s Chemical Reviews paper supplies the historical cathode/anode baseline for lithium-ion cell voltage and energy density (Chem. Rev. 2004, DOI:10.1021/cr020731c, published 2004-09-16, accessed 2025-04-29). Together these two peer-reviewed sources establish that commercial lithium-ion cells operate in the range of roughly 3.6–3.7 V nominal and have cell-level specific energy commonly around 150–250 Wh/kg for NMC/NCA-class chemistries. That range is a review value, not a direct humanoid-robot measurement.
Patent evidence: granted patent vs. pending application
The strongest patent evidence found in this retrieval is a granted U.S. patent, not a pending application:
2.1 Granted patent — hot-swappable robot battery
- Document identifier: U.S. Patent No. US10589409B2
- Title: “Battery replacing device and robot”
- Assignee: UBTECH Robotics Corp.
- Grant date: 2020-03-17
- URL: [patents.google.com
- Access date: 2025-04-29
The patent is a Tier 1 document because it is a granted patent with a grant number, not a patent application. Its claims relate to a battery-replacing device and a robot, i.e., hot-swap/field-replaceable onboard storage. That is a relevant legal record for humanoid-robot energy storage because continuous robot operation often depends on swapping depleted packs rather than waiting for recharge.
Important limits: a granted patent is evidence of recorded claims, not proof of field performance, commercial adoption, or even that the claimed device exists in a shipping robot. The patent does not demonstrate cycle life, capacity, thermal safety, or energy density. Therefore, the patent supports “a party has claimed a hot-swap mechanism” but does not support “a humanoid robot using this patent achieves X hours runtime” (US10589409B2, granted 2020-03-17, accessed 2025-04-29). This is a source-bound distinction: a patent cannot be used as a performance benchmark.
2.2 Peer-reviewed paper evidence
Two independent peer-reviewed battery reviews are used here:
- Dunn, B., Kamath, H., Tarascon, J.-M., “Electrical Energy Storage for the Grid: A Battery of Choices,” Science 334, 928–935 (2011).
DOI: 10.1126/science.1212741, published 2011-11-18, accessed 2025-04-29.
- Whittingham, M.S., “Lithium Batteries and Cathode Materials,” Chemical Reviews 104(10), 4271–4301 (2004).
DOI: 10.1021/cr020731c, published 2004-09-16, accessed 2025-04-29.
Both are peer-reviewed, DOI-verified, and from different institutions/journals. They are general energy-storage reviews, not humanoid-robot-specific performance tests. They anchor the battery-chemistry side of this note. The specific-energy range used in the calculation below is cross-verified between these two reviews, so it is not a [SINGLE-SOURCE] claim.
Company technical disclosures and regulator filings
3.1 Tesla, Inc. — SEC 10-K filing
Tesla’s Form 10-K for fiscal year 2023 is a Tier 1 regulatory filing. It is filed with the U.S. SEC and is a primary source for the company’s own risk disclosures and operating description.
- Document identifier: Tesla, Inc. Form 10-K for fiscal year 2023, Commission file number 001-34756, accession number 0001628280-24-002390
- Filed: 2024-01-29
- URL: [sec.gov
- Access date: 2025-04-29
The 10-K discusses Tesla’s Optimus humanoid-robot program as part of the company’s product-development activities. However, the 10-K does not provide a technical battery-capacity spec for Optimus. It also does not break out separate humanoid-robot revenue or battery cost. That is a disclosure gap, not evidence that the marketed specs are false, but it means an analyst cannot cross-verify Tesla’s Optimus battery marketing claims against Tesla’s SEC filing (Tesla 10-K, accession 0001628280-24-002390, filed 2024-01-29, accessed 2025-04-29).
3.2 Tesla Optimus product page — self-reported battery specs
Tesla’s public Optimus product page lists a battery specification that has been reported as 2.3 kWh and 52 V architecture for the robot. This is a Tier 2 company technical/marketing doc, self-reported, not independently measured, and not repeated in the 10-K.
- URL: [tesla.com
- Access date: 2025-04-29
The 2.3 kWh and 52 V figures should be treated as self-reported product specs. They are not confirmed by Tesla’s SEC filing. This is an example of a marketing/filing discrepancy: public product page gives a number, SEC filing does not. It is not necessarily a contradiction; it is an absence of independent verification (Tesla Optimus page, accessed 2025-04-29).
3.3 Figure AI — Figure 02 pack capacity
Figure AI’s Figure 02 humanoid has a self-reported 2.25 kWh battery pack. This is a company engineering/marketing claim, not an SEC filing or independent benchmark.
- URL: [figure.ai
- Access date: 2025-04-29
This figure is useful as a current-generation pack-size anchor. It is single-source for Figure 02 because the company is private and does not file an SEC 10-K. The 2.25 kWh value is therefore not cross-verified against a regulatory filing; it remains self-reported.
3.4 Unitree H1 — pack capacity
Unitree’s H1 humanoid product page lists a 864 Wh battery capacity. This is another self-reported Tier 2 data point.
- URL: [unitree.com
- Access date: 2025-04-29
3.5 Agility Robotics — Digit runtime
Agility Robotics’ Digit page describes run time/endurance in qualitative or self-reported terms. Any specific “up to 2 hours” or “2–4 hours” runtime claim should be marked needs primary source validation unless confirmed on the page during retrieval.
- URL: [agilityrobotics.com
- Access date: 2025-04-29
Technical architecture of humanoid robot batteries
4.1 Cell chemistry
Current humanoid robots generally use lithium-ion cells with NMC or NCA cathodes. The two independent peer-reviewed reviews above establish the cell-level baseline:
- nominal voltage ~3.6–3.7 V for many lithium-ion chemistries (Whittingham 2004, DOI:10.1021/cr020731c, 2004-09-16, accessed 2025-04-29);
- practical cell-level specific energy ~150–250 Wh/kg for NMC/NCA-class cells (Dunn et al. 2011, DOI:10.1126/science.1212741, 2011-11-18, accessed 2025-04-29).
These are review values, not humanoid-robot-specific measurements. The Tesla Optimus and Figure 02 product pages imply custom lithium-ion packs, but neither identifies exact cell supplier, cathode ratio, or cycle life (Tesla Optimus page, accessed 2025-04-29; Figure 02 blog, accessed 2025-04-29).
4.2 Pack voltage and architecture
Tesla’s self-reported 52 V Optimus pack voltage implies a series-cell lithium-ion architecture, consistent with roughly 14 series cells at 3.7 V nominal (Tesla Optimus page, accessed 2025-04-29). Other humanoid builders do not disclose voltage architecture in accessible primary sources; needs primary source validation.
4.3 Battery management system (BMS)
Every multi-cell lithium-ion pack requires a BMS for state-of-charge, state-of-health, balancing, and safety. This is fundamental to the chemistry, not unique to humanoid robots. The two review papers establish the necessity of managing lithium-ion cycling and safety constraints (Dunn et al. 2011, DOI:10.1126/science.1212741; Whittingham 2004, DOI:10.1021/cr020731c). The UBTECH patent addresses mechanical replacement of a battery, which is one part of operational energy management, not BMS control logic (US10589409B2).
4.4 Thermal management and safety
Humanoid packs are small but can produce high specific heat during fast charge/discharge. The peer-reviewed literature explains that lithium-ion thermal runaway, electrode stability, and flammable electrolyte are inherent risks, so pack design must include thermal barriers, current interruption, and possibly liquid or air cooling (Dunn et al. 2011, DOI:10.1126/science.1212741; Whittingham 2004, DOI:10.1021/cr020731c). None of the company sources above discloses a humanoid-specific thermal design; needs primary source validation.
4.5 Hot-swappable and field-replaceable storage
The granted UBTECH patent claims a battery-replacing device and robot, which is direct evidence of a claimed hot-swap/field-replacement solution (US10589409B2, granted 2020-03-17, accessed 2025-04-29). This is relevant to high-utilization warehouse/factory robots that cannot be taken offline for long recharging. Again, patent is evidence of a claim, not evidence that a shipping robot uses that mechanism.
Worked quantitative calculation: runtime and battery mass
The most useful desktop calculation is to size a pack from a robot’s mission-average power and then invert it to compute runtime for the self-reported packs above.
5.1 Formula
Runtime formula:
\[
t_{\text{operate}} = \frac{E_{\text{pack}} \times \text{DoD}}{P_{\text{avg}}}
\]
where:
- \(E_{\text{pack}}\) = pack energy capacity (Wh);
- \(\text{DoD}\) = usable depth of discharge, assumed 0.9 for illustrative lithium-ion pack operation;
- \(P_{\text{avg}}\) = mission-average electrical power draw (W).
Battery mass formula:
\[
m_{\text{battery}} = \frac{P_{\text{avg}} \times T}{e_{\text{cell}} \times \eta_{\text{pack}}}
\]
where:
- \(e_{\text{cell}}\) = cell-level specific energy (Wh/kg);
- \(\eta_{\text{pack}}\) = cells-to-pack mass efficiency;
- \(T\) = desired runtime.
5.2 Inputs
| Input | Value | Source | Character |
|---|---|---|---|
| Cell-level specific energy \(e_{\text{cell}}\) | 250 Wh/kg | Dunn et al. 2011, DOI:10.1126/science.1212741, accessed 2025-04-29 | Review value, upper practical bound; cross-verified by Whittingham 2004 |
| Cells-to-pack mass efficiency \(\eta_{\text{pack}}\) | 0.65 | Assumed | Assumed, not retrieved |
| Mission-average power \(P_{\text{avg}}\) | 400 W | Assumed | Assumed/inferred; needs primary source validation for any specific robot |
| Usable depth of discharge DoD | 0.9 | Assumed | Assumed |
| Figure 02 pack energy | 2250 Wh | Figure 02 blog, 2024-08-06, accessed 2025-04-29 | Self-reported |
| Optimus pack energy | 2300 Wh | Tesla Optimus page, accessed 2025-04-29 | Self-reported |
| Unitree H1 pack energy | 864 Wh | Unitree H1 page, accessed 2025-04-29 | Self-reported |
5.3 Calculation 1: runtime at assumed 400 W average load
For Figure 02:
\[
t = \frac{2250 \times 0.9}{400} = 5.06\ \text{hours}
\]
For Tesla Optimus:
\[
t = \frac{2300 \times 0.9}{400} = 5.18\ \text{hours}
\]
For Unitree H1:
\[
t = \frac{864 \times 0.9}{400} = 1.94\ \text{hours}
\]
Sensitivity:
- At \(P_{\text{avg}} = 250\) W, Figure 02 runtime ≈ \(2250 \times 0.9 / 250 = 8.1\) hours.
- At \(P_{\text{avg}} = 600\) W, Figure 02 runtime ≈ \(2250 \times 0.9 / 600 = 3.4\) hours.
- At DoD = 0.8, Figure 02 runtime at 400 W ≈ \(2250 \times 0.8 / 400 = 4.5\) hours.
This calculation is illustrative. The 400 W average power is not an independent measurement; it must be labeled needs primary source validation. If a real measured power draw is lower, runtime grows; if a robot spends significant time climbing stairs, lifting payloads, or recovering from disturbances, average power can rise well above 400 W.
5.4 Calculation 2: battery mass for an 8-hour shift
\[
m_{\text{battery}} = \frac{400 \times 8}{250 \times 0.65} = \frac{3200}{162.5} = 19.7\ \text{kg}
\]
Sensitivity:
| Power | Mass |
|---|---|
| 250 W | 12.3 kg |
| 400 W | 19.7 kg |
| 600 W | 29.5 kg |
| 1000 W | 49.2 kg |
At current cell-level 250 Wh/kg and 65% pack efficiency, a humanoid robot with a 400 W mission-average draw would need roughly 20 kg of battery for an eight-hour shift. That is a large fraction of most humanoid robot masses. The assumed pack efficiency and power draw are uncorroborated, so the mass output is directional only. The load-bearing technical inputs are the cell-level specific energy from the two peer-reviewed reviews; the 0.65 pack efficiency and 400 W load are assumptions, not established facts.
Parameter table
| Parameter | Value | Source (URL + ID + date) | Tier | measured/derived/assumed/inferred | cross-verified? |
|---|---|---|---|---|---|
| Li-ion cell specific energy | ~150–250 Wh/kg | Dunn et al. 2011, DOI:10.1126/science.1212741, 2011-11-18, access 2025-04-29 | Tier 1 | review value | yes, with Whittingham 2004, DOI:10.1021/cr020731c |
| Li-ion nominal voltage | ~3.6–3.7 V | Whittingham 2004, DOI:10.1021/cr020731c, 2004-09-16, access 2025-04-29 | Tier 1 | review value | single-source |
| Tesla Optimus pack capacity | 2.3 kWh, 52 V | Tesla Optimus page, access 2025-04-29 | Tier 2 | self-reported product spec | no |
| Figure 02 pack capacity | 2.25 kWh | Figure 02 blog, 2024-08-06, access 2025-04-29 | Tier 2 | self-reported product spec | no |
| Unitree H1 pack capacity | 864 Wh | Unitree H1 page, access 2025-04-29 | Tier 2 | self-reported product spec | no |
| Hot-swap battery replacement claim | One granted patent | US10589409B2, granted 2020-03-17, access 2025-04-29 | Tier 1 | legal claim | single-source |
| Mission-average robot electrical power | 400 W used for calculation | Assumption; needs primary source validation | n/a | assumed/inferred | no |
| Usable depth of discharge | 90% used for calculation | Assumption | n/a | assumed | no |
| Cells-to-pack mass efficiency | 65% used for calculation | Assumption | n/a | assumed | no |
| Humanoid robot pack-level cycle life | needs primary source validation | No retrieved primary source | n/a | unknown | no |
The only cross-verified quantitative technical parameter in this table is the cell-level specific-energy range, because two independent peer-reviewed reviews from different journals support it. The specific humanoid pack capacities are all self-reported company numbers and are not independently tested here.
Contradictions and source-bound gaps
7.1 SEC filing vs. marketing spec
Tesla’s marketing/product page gives Optimus battery capacity and voltage, but Tesla’s 10-K does not repeat and therefore does not independently verify those numbers (Tesla 10-K, filed 2024-01-29, accessed 2025-04-29; Tesla Optimus page, accessed 2025-04-29). This is not proven fraud; it is a verification gap: the SEC filing is silent on pack specs, so the 2.3 kWh/52 V claim is single-source from the company.
7.2 Patent vs. performance
The granted patent US10589409B2 is often misinterpreted as proof that a robot’s hot-swap system works economically. It is a legal document recording claims, not a test report. The note therefore treats it as evidence of a moat/legal position only (US10589409B2, granted 2020-03-17, accessed 2025-04-29).
7.3 Company self-reported pack sizes vary widely
Company self-reported pack capacities vary from Unitree H1’s 0.864 kWh (Unitree H1 page, accessed 2025-04-29) to Figure 02’s 2.25 kWh (Figure 02 blog, accessed 2025-04-29) and Tesla Optimus’ 2.3 kWh (Tesla Optimus page, accessed 2025-04-29). Such variance may reflect different design missions, degrees of freedom, actuator efficiency, or advertised endurance, but without independent power-draw measurements, runtime cannot be fairly compared.
7.4 No independent humanoid battery benchmark located
No independent humanoid-robot battery benchmark or regulator technical assessment was located in this retrieval. Therefore, every specific pack capacity and runtime figure in this note carries the weaker Tier 2 self-reported status or an explicit needs primary source validation.
Unit economics / business model
The value chain monetizes humanoid-robot energy storage at multiple layers:
- Battery cell/pack suppliers sell cells and/or complete packs to robot OEMs. Margins are not disclosed in the retrieved humanoid-specific sources; pack-level margins for industrial batteries generally depend on volume and cathode/cell chemistry.
needs primary source validationfor any specific number such as $/kWh or gross margin. - Robot integrators monetize the robot itself, sometimes as hardware plus software, or as a service/leasing arrangement. Tesla’s 10-K does not break out Optimus revenue or robot battery cost, so robot-level unit economics are not yet measurable from SEC filings (Tesla 10-K, filed 2024-01-29, accessed 2025-04-29).
- Potential battery-swap/field-service business model: a hot-swappable pack could generate recurring service revenue from swappable battery depots. The UBTECH granted patent records a hot-swap enabling mechanism, but no commercial swap-service financials were located (US10589409B2, granted 2020-03-17, accessed 2025-04-29).
- Value capture is not yet proven: because Tesla’s SEC filing does not separate Optimus revenue or battery cost, and Figure AI is private, the current business model is mostly pre-revenue or non-disclosed for humanoid-robot energy storage specifically.
No fabricated financial numbers are included. Any specific $/kWh, cycle-cost, or margin figure would need primary-source validation.
Disconfirming Evidence
The thesis implied by this note is that onboard lithium-ion energy storage is the near-term enabling constraint for humanoid robots, with company-specific pack sizes in the 0.8–2.3 kWh range and possible hot-swap/battery-swap utility.
This thesis would be wrong or materially weakened if:
- an independent, institution-neutral benchmark measures a shipping humanoid robot’s pack-level specific energy or cycle life far below the 150–250 Wh/kg review range or materially below the company’s self-reported capacity;
- a regulator or mandatory safety filing reports thermal runaway, repeated pack field failures, or a recall in a humanoid robot battery pack;
- the granted patent US10589409B2 is narrowed, invalidated, or abandoned in a way that removes the hot-swap/field-replacement claim;
- a company’s SEC filing discloses that humanoid robot battery cost, failure rate, or recharge downtime makes the robot uneconomic relative to human labor or conventional automation; or
- named customer contracts involving humanoid robots are canceled or fail to scale, indicating that onboard energy storage/charging infrastructure is not meeting operational requirements.