Independent reliability intelligence
We track what real deployments disclose about runtime, failures, human intervention and recovery — and make the missing data visible.
01 / Reliability records
A successful task is not the same as a reliable system. Each record separates observed output from vendor claims and unknowns — without inferring uptime from incomplete data.
Automotive production · Spartanburg, USA · Completed
BMW and Figure report substantial Figure 02 factory runtime and output, while the scheduled-time, downtime and intervention data needed to calculate availability remain unpublished.
Warehouse logistics · Flowery Branch, Georgia, USA · Commercial
Digit performs a commercial tote-moving workflow at GXO, but the published duration and output milestones do not reveal operating availability or human-support frequency.
Industrial humanoid · Product-level claim · No operator dataset reviewed
Walker S2's battery architecture is presented as enabling continuous operation, but no bounded field result demonstrates measured availability.
Automotive logistics · Spartanburg, USA · Announced / demonstration stage
BMW confirms a Figure 03 logistics project and Figure shows the intended workflow, but no sustained production period or quantified operating result is public.
Controlled laboratory benchmark · Stuttgart, Germany · Single-unit test
Fraunhofer independently benchmarked a specified Unitree G1 configuration, producing laboratory measurements that characterize one test unit rather than field uptime or fleet reliability.
Autonomous tote handling in electronics-factory logistics · Erlangen, Germany · Completed proof of concept
Siemens confirms a two-week HMND 01 factory test with throughput and task-success figures, but the published uptime wording lacks the denominator required for availability.
Multi-task retail pilot · Mark's store, Langley, British Columbia, Canada · Completed one-week pilot
Sanctuary reports a one-week retail pilot and 110 correctly completed tasks, without publishing the attempts, operating time or control mode needed to assess reliability.
Heavy-component transport around an industrial washing process · Cheraw, South Carolina, USA · Active factory use
Schaeffler describes Digit performing a defined Cheraw factory task more than a year apart, without evidence of continuous operation or published reliability rates during that period.
Battery assembly and component-production pilot · Leipzig, Germany · Active factory pilot / production tasks reported
BMW confirms a staged AEON factory pilot and Hexagon reports the start of production tasks, while runtime, task output, interventions and recovery remain unpublished.
Intralogistics training and production-environment data collection · Berlin-Marienfelde, Germany · Production-environment training / autonomy development
Mercedes-Benz confirms Apollo training and data collection in a production environment, but does not publish autonomous task results, operating time or reliability measurements.
Automotive parts-sequencing pilot · Hyundai Motor Group Metaplant America, Georgia, USA · Completed prototype pilot / product deployment planned
Hyundai confirms a completed Atlas prototype pilot for automotive parts sequencing, without publishing duration, output, intervention or reliability measurements.
Near-production bearing-ring bin-picking proof of concept · Schaeffler site, Erlangen, Germany · Completed near-production PoC / successor deployment planned
Schaeffler confirms a completed pre-alpha proof of concept for bearing-ring handling, but no quantitative operating, task-success or reliability result is public.
Year-long, multi-phase logistics integration pilot · ROSSMANN logistics, Burgwedel, Germany · Active multi-phase logistics pilot
ROSSMANN confirms one Walker S2 in a phased logistics pilot, while the reported task activity has no public runtime, output or reliability measurements.
02 / Compare evidence
Status reflects public disclosure in the reviewed source set — not the robot's actual performance.
| Record | Runtime hours | Availability / uptime | Human interventions | MTBF | MTTR | Task output | Energy / charging | Autonomy mode | Fleet size |
|---|---|---|---|---|---|---|---|---|---|
| Figure 02× BMW | Operator corroborated: 1,250+ hours | Not disclosed: Not disclosed | Not disclosed: Logged, not published | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Operator corroborated: 90,000+ parts | Not disclosed: Not disclosed | Manufacturer reported: Claimed; share unknown | Not disclosed: Not disclosed |
| Digit× GXO | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Manufacturer reported: 100,000+ totes | Not disclosed: Not disclosed | Manufacturer reported: Asserted; share unknown | Not disclosed: Not disclosed |
| Walker S2continuous-operation claim | Not disclosed: Not disclosed | Capability claim: 24/7 claim only | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Manufacturer reported: 3 min swap claim | Manufacturer reported: Maker reported | Not disclosed: Not disclosed |
| Figure 03× BMW | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Manufacturer reported: Maker reported | Not disclosed: Not disclosed |
| Unitree G1 EDU-4× Fraunhofer IPA | Not disclosed: Total test hours not disclosed | Not applicable to this record: Not applicable to this lab benchmark | Not disclosed: Not disclosed | Not disclosed: Not measured | Not disclosed: Not measured | Not disclosed: No production task | Independently measured: 1:49 scenario / 239 W | Independently measured: Teleoperated baseline | Independently measured: 1 test unit |
| HMND 01 Alpha× Siemens | Operator corroborated: >8 h reported; scope unclear | Not disclosed: No scheduled-time denominator | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Operator corroborated: 60 moves/h; >90% autonomous | Not disclosed: Not disclosed for POC | Operator corroborated: Autonomous task execution | Operator corroborated: 1 robot described |
| Unnamed Sanctuary GPR× Canadian Tire / Mark's | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Manufacturer reported: 110 tasks reported correct | Not disclosed: Not disclosed | Not disclosed: Pilot mode not disclosed | Manufacturer reported: 1 robot described |
| Digit× Schaeffler | Not disclosed: No cumulative operating hours | Not disclosed: No scheduled-time denominator | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: No cycle count or success rate | Not disclosed: No charge/runtime record | Not disclosed: Task-level mode not disclosed | Operator corroborated: 1 site unit described; total unknown |
| AEON× BMW | Not disclosed: No cumulative operating hours | Not disclosed: No scheduled-time denominator | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: No task count or success rate | Capability claim: Up to 4 h / auto-swap product claim | Not disclosed: Current autonomy share not disclosed | Not disclosed: Not disclosed |
| Apollo× Mercedes-Benz | Not disclosed: No cumulative operating hours | Not disclosed: No scheduled-time denominator | Operator corroborated: Human teaching disclosed; intervention rate unknown | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: No task count or success rate | Not disclosed: No charge/runtime record | Operator corroborated: Teleoperation used; autonomy training underway | Operator corroborated: Multiple robots described; exact count unknown |
| Atlas prototype× Hyundai Motor Group | Not disclosed: No duration or operating hours | Not disclosed: No scheduled-time denominator | Manufacturer reported: Teleoperated training described; pilot intervention rate unknown | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Repeated sequencing described; no count or success rate | Not disclosed: Not disclosed for the tested prototype | Operator corroborated: Autonomous operation stated; share and protocol unknown | Not disclosed: Not disclosed for the 2025 pilot |
| Humanoid pre-alpha robot× Schaeffler | Not disclosed: No test duration or operating hours | Not disclosed: No scheduled-time denominator | Manufacturer reported: Teleoperation used for training; test interventions unknown | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: No ring count, attempts or throughput | Not disclosed: Not disclosed | Manufacturer reported: Recovery autonomy claimed; test mode unknown | Manufacturer reported: One pre-alpha robot described |
| Walker S2× ROSSMANN | Not disclosed: 2026 pilot plan; no operating hours | Not disclosed: No scheduled-time denominator | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Not disclosed: Not disclosed | Manufacturer reported: Material handling and sorting reported; no counts | Capability claim: Self-changing battery claimed; no observed duty cycle | Capability claim: Battery-swap capability stated; task autonomy unknown | Operator corroborated: 1 Walker S2 confirmed by ROSSMANN |
Missing is a finding. “Not disclosed” means no qualifying public value was identified. “Not applicable” is used only when a field does not fit the bounded record and includes a reason on the record page.
See what can—and cannot—be comparedDeployment evidence
See field programs, pilots, announced deployments and research evidence without treating them as equivalent.
Explore deployment evidenceEvidence log
New measurements, sources and material corrections — separated from general robotics news and linked to the record they change.
Records, claims, disclosures, sources, history, deployments, coverage and metric definitions now have versioned CSV downloads, stable joins, a manifest and per-file checksums.
The Figure 02 × BMW and Unitree G1 × Fraunhofer IPA pilot adds exact locators, capture metadata and SHA-256 fingerprints while keeping the third-party source copies outside the public site.
The initial universe now shows 13 full records, 16 deployment signals and one no-result outcome, alongside search limits and a versioned downloadable dataset. Private candidate observations remain private.
Every record now separates who confirms the deployment or test from who supports the available operating results. Ambiguous overall labels such as ‘Mixed public record’ have been removed.
03 / Methodology
Every material statement should remain traceable to a dated source and preserve the conditions under which it was observed.
Every deployment, metric and claim links back to its publisher.
Operator corroboration, manufacturer reporting and independent measurement are not interchangeable.
Unknown stays unknown. Missing data is never replaced by an estimate or zero.
Open record
Operators, researchers and manufacturers can help make humanoid reliability measurable.
Submit evidence