Fraunhofer independently measured one specified Unitree G1 for power use, battery endurance, walking, stability, collision force, temperature limits and security.
HU-R005 / Independent benchmark
Unitree G1 EDU-4 × Fraunhofer IPA
Deployment / test evidenceIndependent laboratory benchmark
Reliability evidenceIndependent lab measurements; no field uptime
Fraunhofer independently benchmarked a specified Unitree G1 configuration, producing laboratory measurements that characterize one test unit rather than field uptime or fleet reliability.
1:49 hderived typical-scenario endurance
239 Wmean scenario power
>500 Nreported collision force
Cite this recordHU-R005 · rev. 12 · Public Data v1.4.0
HumanoidUptime. “Unitree G1 EDU-4 × Fraunhofer IPA” (HU-R005), rev. 12, evidence current to 2026-09-04. HumanoidUptime Public Data v1.4.0 (2026). https://humanoiduptime.com/id/HU-R005
- Persistent identity
- HU-R005 · revision 12
- Current record page
- https://humanoiduptime.com/records/unitree-g1-fraunhofer-ipa/
- Evidence cutoff
- 2026-09-04
- Dataset release
- v1.4.0 · snapshot 2026-09-04
- Immutable dataset
- Versioned JSON
- JSON SHA-256
0e174ef9f5d6de3f140478daa628032f4610761ba3794ec124bd310b33ec9705
What this means
The short version, without reliability jargon
The benchmark has no availability ratio because it defines no scheduled operating window. It also does not show field availability, fleet performance, failure frequency, human help or recovery time.
The benchmark provides valuable objective measurements of one test unit, but laboratory results cannot be treated as proof of uptime in a real deployment.
Evidence verdict
Established. Claimed. Unknown.
EstablishedOperator or independent evidence
- Test configurationG1 EDU-4, Dex3-1 three-finger hands, firmware 1.04[3] [4]
- Energy and endurance239 W mean power and 1:49 h derived per-charge endurance in the defined typical scenario[4]
- Walking speed0.49 m/s slow and 0.84 m/s fast walking measured with 3D tracking[4]
- Thermal constraintOutstretched arms could shut down after one to two minutes without added payload[4]
- Collision forceMean peak forces of approximately 510 N at the chest and 530 N at the abdomen in specified walking collisions[1] [2] [3]
- Stability and recoveryNo balance loss on cable-channel steps or a 20% ramp; recovery depended on floor friction[4]
- CybersecurityBluetooth remote-code-execution vulnerability found in the tested software state[3] [4]
- Delivered autonomyRemote-controlled walking available after delivery; further capabilities require user implementation[4]
ClaimedManufacturer or capability claim
No separate manufacturer-only claim is material to this record.
Still unknownNo qualifying public value
- Total benchmark hours, repeat counts and uncertainty
- Failure and human intervention rate
- MTBF and MTTR
- Availability in a scheduled field deployment
What the evidence says
Fraunhofer publishes useful measurements that reveal concrete operating constraints, including scenario-based power draw, derived per-charge endurance, actuator overheating and collision force. Availability is not applicable to this bounded laboratory benchmark because it defines no scheduled operating window; the tests also do not establish field availability, failure frequency, intervention rate or repair time. The public material does not report repeat counts or uncertainty for most tests.
Analysis scope / Mission Model v1
What exactly was evaluated
One specified Unitree G1 EDU-4 test unit
One identified or described robot
Sources: [3] [4]- TaskKnown
- Separate laboratory scenarios for energy, movement, stability, safety, cleanliness and securitySources: [3] [4]
- Success definitionPartially defined
- Test conditions are described, but repeat counts and uncertainty are absent for most scenariosSources: [4] [1]
- Object or payloadPartially defined
- A 3 kg walking load is reported for one scenario; payload varies or is not relevant elsewhereSources: [4]
- Operating scheduleNot applicable
- The benchmark uses separate test scenarios rather than a scheduled operating shiftA production schedule is outside the bounded laboratory benchmark.
- Energy strategyKnown
- Scenario power draw is measured and per-charge endurance is derived; charging downtime is not testedSources: [4]
- Control modePartially defined
- Remote-controlled walking is the delivered baseline; additional autonomy requires user implementationSources: [4]
- Human-support boundaryPartially defined
- Test personnel execute controlled scenarios, but support and reset time are not reported as a metricSources: [4]
Known describes the bounded record—not every configuration or deployment of this robot. Partially defined fields retain the published context while showing the missing boundary.
01 / Reliability fields
What is known—and missing
Runtime / operating hoursTime accumulated while a robot is operating under the source's stated rules. It may be summed across several robots and does not necessarily include pauses or downtime.Learn more →
Not disclosed: Total test hours not disclosed
Availability / uptimeThe share of a defined scheduled period in which a system was able to perform the required function. The time boundary, required function and rules for counting downtime must be stated.Learn more →
Not applicable to this record: Not applicable to this lab benchmarkThe benchmark consists of separate laboratory scenarios and defines no scheduled operating window. This applies only to the bounded test record; availability in a field deployment remains unknown.
Human interventionA human action needed to keep, restore or redirect the robot's work, such as a reset, remote correction or physical assistance.Learn more →
Not disclosed: Not disclosed
MTBFMean time between failures: the average operating time between one defined failure and the next in a repairable system.Learn more →
Not disclosed: Not measured
MTTRAn average time associated with restoring operation after a failure. The acronym is used for repair, restoration or recovery in different contexts, so the included steps must be defined.Learn more →
Not disclosed: Not measured
Task outputThe number of reported items moved, parts handled, cycles completed or other units produced.Learn more →
Not disclosed: No production task
Energy / charging
Independently measured: 1:49 scenario / 239 W
Autonomous operationTask execution by the robot without a person continuously issuing the individual control actions. The system may still be supervised and may request help.Learn more →
Independently measured: Teleoperated baseline
Fleet sizeThe number of robots included in a reported result or observation.Learn more →
Independently measured: 1 test unit
Independently measured
Operator corroborated
Manufacturer reported
Capability claim
Not disclosed
Not applicable to this record
02 / Claims and evidence
What each source supports
Test configuration
Independently measured
G1 EDU-4, Dex3-1 three-finger hands, firmware 1.04
Fraunhofer identifies the tested configuration and states that the unit was delivered in May 2025; results should not be generalized to other variants or firmware versions without qualification.
Sources: [3] [4]
Energy and endurance
Independently measured
239 W mean power and 1:49 h derived per-charge endurance in the defined typical scenario
Fraunhofer measures power in standardized standing and walking conditions and derives maximum operating times from those results. The 1:49 figure is not presented as an observed continuous field run or cumulative runtime. Fraunhofer also reports 154 W standing, 272 W level walking, 283 W at 10% incline and a derived 2:49 h for standing only.
Metric interpretationDifferent metric — not availability
- Source wording
- maximum operating time in the typical scenario · 1:49 h per charge
- Tested against
- Technical availability
- Normalized result
- Not calculated
This is a scenario-derived energy-endurance result. It is valid for the stated laboratory scenario, but it does not measure availability against scheduled operating time.
- Observation windowDefined laboratory power-use scenario
- Scheduled operating timeNo scheduled operating window belongs to this energy calculation
- Technical downtimeThe calculation models energy endurance, not service interruptions
- Availability exclusionsNo availability ratio is calculated
Walking speed
Independently measured
0.49 m/s slow and 0.84 m/s fast walking measured with 3D tracking
A 3 kg load did not reduce steady walking speed in the described test, but increased acceleration and deceleration time by several tenths of a second.
Sources: [4]
Thermal constraint
Independently measured
Outstretched arms could shut down after one to two minutes without added payload
Fraunhofer attributes the shutdown and dropping arms to actuator overheating. No frequency across repeated trials or recovery time is published.
Sources: [4]
Collision force
Independently measured
Mean peak forces of approximately 510 N at the chest and 530 N at the abdomen in specified walking collisions
Fraunhofer's public safety guide states that displayed values are mean peak forces from at least three trials per scenario with the robot in fast mode at about 1 m/s. It does not publish the underlying observations, spread or unit-to-unit variation.
Sources: [1] [2] [3]
Stability and recovery
Independently measured
No balance loss on cable-channel steps or a 20% ramp; recovery depended on floor friction
Self-stabilization was strong in the described tests, while standing up from the back failed on smooth tile and wood because the arms could slip.
Sources: [4]
Cybersecurity
Independently measured
Bluetooth remote-code-execution vulnerability found in the tested software state
Fraunhofer reports that the vulnerability persisted across several firmware updates and was subsequently closed by a newer update.
Sources: [3] [4]
Delivered autonomy
Independently measured
Remote-controlled walking available after delivery; further capabilities require user implementation
This is the observed baseline of the tested EDU configuration, not a claim about every later software stack built on the platform.
Sources: [4]03 / Evidence over time
How the public record changed
This is a history of public knowledge, not proof that robot performance changed on these dates. Earlier states were reconstructed later from dated sources; the reconstruction date remains visible.
Independent safety measurements become public
Fraunhofer IPA published collision-force measurements for one Unitree G1 test unit and described the test repetitions and measurement setup. Energy endurance, field runtime, availability and failure-recovery metrics were not part of this public state.
1/9 applicable fields disclosedCoverage at this knowledge date—not performance.
- Fleet size1 test unit
Reconstruction basis and limits
Reconstructed solely from Fraunhofer IPA's dated October 2025 safety guide.
- Fraunhofer identified one Unitree G1 test robot with Dex3-1 hands.
- The guide reported average peak forces from at least three trials per collision scenario.
- Forward walking collisions were reported at approximately 510 N at the chest and 530 N at the abdomen.
- The guide labels the configuration G1 EDU+; the later benchmark publication specifies EDU-4 and firmware 1.04.
- Safety-test results are not deployment uptime or field reliability measurements.
The broader independent benchmark is announced
Fraunhofer specified the tested EDU-4 configuration and added independently derived per-charge endurance, cleanroom, stability and cybersecurity findings. The typical energy scenario reached 1:49 hours, but this was not a continuous field run or measured availability.
2/9 applicable fields disclosedCoverage at this knowledge date—not performance.
- Runtime hoursNot disclosed →No field runtime disclosed
- Energy / chargingNot disclosed →1:49 typical / 2:49 standing
- Autonomy modeTest operating mode not quantified →Delivered operating mode not stated
Reconstruction basis and limits
Reconstructed cumulatively from the dated safety material and Fraunhofer IPA's 27 May 2026 benchmark press release.
- Fraunhofer specified G1 EDU-4, Dex3-1 hands and firmware 1.04.
- Fraunhofer reported 1:49 hours for a typical standing-and-walking scenario and 2:49 hours while standing.
- The release added independently assessed collision, stability, cleanroom and cybersecurity findings.
- The endurance figures were scenario-derived single-charge results, not cumulative operating hours.
- No scheduled time, downtime, MTBF, MTTR or intervention rate was published.
Detailed measurements expose the operating boundary
Fraunhofer's detailed publication added mean power, walking speed, thermal, stability and recovery context and clarified that remote-controlled walking was the delivered baseline. The evidence became deeper without becoming a field-uptime record.
3/9 applicable fields disclosedCoverage at this knowledge date—not performance.
- Energy / charging1:49 typical / 2:49 standing →1:49 scenario / 239 W
- Autonomy modeDelivered operating mode not stated →Teleoperated baseline
Reconstruction basis and limits
Reconstructed cumulatively from all dated Fraunhofer sources available through 3 June 2026.
- Mean power in the defined typical scenario was 239 W.
- Slow and fast walking were measured at 0.49 m/s and 0.84 m/s using 3D tracking.
- The delivered baseline supported remote-controlled walking; further capabilities required user implementation.
- Laboratory measurements should not be generalized into field availability or fleet reliability.
- The results apply to the identified hardware and software state.
Current public evidence state
Fraunhofer publishes useful measurements that reveal concrete operating constraints, including scenario-based power draw, derived per-charge endurance, actuator overheating and collision force. Availability is not applicable to this bounded laboratory benchmark because it defines no scheduled operating window; the tests also do not establish field availability, failure frequency, intervention rate or repair time. The public material does not report repeat counts or uncertainty for most tests.
3/8 applicable fields disclosed · 1 not applicableCoverage at the latest review—not performance.
- Runtime hoursNo field runtime disclosed →Total test hours not disclosed
- Availability / uptimeNot disclosed →Not applicable to this lab benchmark
- MTBFNot disclosed →Not measured
- MTTRNot disclosed →Not measured
- Task outputNot disclosed →No production task
04 / Deployment timeline
What happened, and when
-
Safety findings described
Fraunhofer discusses collision-force testing of the G1 and the implications for operation near people.
[2] -
Standardized benchmark announced
Fraunhofer publishes the benchmark framework and selected G1 results across six test categories.
[3] -
Detailed measurements published
The extended account adds energy, speed, stability, thermal, cleanliness and cybersecurity observations.
[4]
05 / Sources
Trace every finding
-
Source 1Independent research institute
Leitfaden: Funktionale Sicherheit humanoider Roboter ↗
Fraunhofer IPA · Published 2025-10-25 · Accessed 2026-09-04
Preserved source · HU-SA-R005-001PDF page 9, ‘4. Sicherheitsrisiken: Kollisionen mit dem laufenden Roboter’.SHA-256 11ad99c755c1735e8a12c66fe5654440899e8108ec575e8b810dcdfce08dedb3Captured 2026-09-04T14:10:51Z · application/pdf · 2,148,565 bytes · Internal verification copy, not publicly redistributed -
Source 2Independent research institute
Humanoide Roboter – Status quo, Potenziale und Forschungstätigkeiten (Teil 3 von 3) ↗
Fraunhofer IPA · Published 2026-02-12 · Accessed 2026-09-04
Preserved source · HU-SA-R005-002Section ‘Sicherheit’, opening two paragraphs on collision testing and measured force.SHA-256 159d49ae2c354ac19ac36ee1ad3377e328ab26e8e3c5aa0ea3cd24cb81372045Captured 2026-09-04T14:10:52Z · text/html; charset=UTF-8 · 258,765 bytes · Internal verification copy, not publicly redistributed -
Source 3Independent research institute
Fraunhofer IPA develops standardized analyses for application-relevant criteria of humanoid robots ↗
Fraunhofer IPA · Published 2026-05-27 · Accessed 2026-09-04
Preserved source · HU-SA-R005-003‘What is the benchmark?’ section, six-category list and following Unitree G1 configuration and result paragraphs.SHA-256 4f546b280b72d5e87ed7a62eafa0e0257b47cf522ab005cd05eeaff744f985bdCaptured 2026-09-04T14:10:52Z · text/html;charset=utf-8 · 41,319 bytes · Internal verification copy, not publicly redistributed -
Source 4Independent research institute
Humanoide Roboter: Standardisierte Analysen für anwendungsrelevante Kriterien ↗
Fraunhofer IPA · Published 2026-06-03 · Accessed 2026-09-04
Preserved source · HU-SA-R005-004Sections 1 ‘Technologien und Basisfähigkeiten’, 2 ‘Komplexe Fähigkeiten’, 4 ‘Funktionale Sicherheit’, 5 ‘Cybersicherheit’ and 6 ‘Energieeffizienz’.SHA-256 2f5acfe1de0f96cd25a9de8fb0cb8cced3cbc979091d9e37170792d89562872aCaptured 2026-09-04T14:10:58Z · text/html; charset=UTF-8 · 262,354 bytes · Internal verification copy, not publicly redistributed