OEM and supplier buyer guide
Start with the task, not the body shape. Compare humanoids with cobots, AMRs and fixed automation, then make every pilot earn its place on takt, quality, safety, integration and service.
The form-factor decision
01
A plausible humanoid case is a variable, human-designed workflow that cannot be simplified into transport plus a fixed handling station. Start with one repeatable task, not a promise of general-purpose labor.
02
Fixed automation remains the default for stable high-throughput work. Cobots suit bounded shared stations. AMRs move material on repeatable routes. Each removes complexity the humanoid would otherwise have to carry.
03
A flexible form factor has value only if it reduces retooling, covers multiple qualified tasks or reaches work that simpler systems cannot—without sacrificing takt, safety, quality or supportability.
Automotive task decision matrix
These are screening rules, not universal recommendations. The exact part, station, route, process capability and safety assessment must decide the application.
Swipe to compare all four platform classes →
| Task | HumanoidMobile manipulation | CobotBounded shared station | AMRTransport | Fixed automationStable high throughput |
|---|---|---|---|---|
| Component sequencing | Consider when unsorted parts must be picked into a sequence trolley inside a human-designed logistics area; BMW is starting this workflow with Figure 03.[BMW1] | Prefer when the pick-and-place step stays inside one bounded station and part presentation can be controlled.[FAN1] | Use for repeatable transport after sequencing. BMW describes a tugger or Smart Transport Robot carrying the prepared trolley onward.[BMW1] | An established option for stable, high-throughput picking when part presentation, reach and downstream placement are repeatable.[FAN1] |
| Line-side logistics | Use only when mobility and manipulation are both required and the task cannot be separated into a transport step and a fixed handling step.[BMW1][MB1] | Fit for assisted loading or presentation at a defined line-side station, subject to the application safety case.[FAN1][ISO2][ISO3] | Use for recurring tote, rack or cart movement on mapped routes where manipulation is not part of the mission.[BMW1] | Use conveyors, tuggers or fixed transfer when flow and destinations are stable enough to engineer once.[BMW1][FAN1] |
| Material handling | Pilot for mixed, human-scale handling across changing workstations. BMW’s completed Figure 02 pilot establishes one sheet-metal positioning case, not universal fit.[BMW1][BMW2] | Prefer for bounded, lower-payload handling where a human shares the process and the workstation can be integrated around the application.[FAN1][ISO3] | Prefer for moving material between stations without grasping, orienting or placing individual parts.[BMW1] | An established option for fast, repeatable transfer, heavy payloads or a mature process with predictable geometry.[FAN1] |
| Bin picking | Consider when the robot must reposition its body, manage mixed presentation and complete a downstream mobile step.[BMW1] | Prefer for lower-rate, bounded picks when the bin, vision and destination remain at one station.[FAN1] | An AMR moves the bin; it does not replace the picking system unless a supported manipulator is integrated.[BMW1] | An established option when vision-guided picking can be engineered around a stable bin, part family and placement target.[FAN1] |
| Machine tending | Roadmap candidate where several human-designed machines must be served by one mobile platform. Hyundai lists machine tending, but not as a proven production deployment.[HMG1] | Strong fit for one machine or a compact cell when loading, controls and guarding can be integrated around a repeatable cycle.[FAN1][ISO2] | Useful for workpiece transport between machines; manipulation requires a separate top module or robot cell.[FAN1] | An established option for high utilization on a stable machine, part family and loading geometry.[FAN1] |
| Inspection | Consider only when inspection must be combined with handling or tool use. Mercedes is testing initial component quality checks with Apollo.[MB1] | Prefer when a sensor or gauge must be presented repeatedly to a part at a bounded workstation.[FAN1] | Prefer for flat, repeatable routes when a validated inspection payload can be integrated without manipulation.[BMW1][FAN1] | An established option for repeatable in-line visual, acoustic or metrology checks; BMW already describes fixed camera and sensor AI for quality inspection.[BMW1][FAN1] |
| Kitting | Candidate when one system must pick varied items, verify them and move between kit locations built for people.[MB1][BMW1] | Prefer when kit locations can be consolidated at a bounded station with controlled item presentation.[FAN1] | Prefer for moving complete kits or presenting racks; use a separate picking system for item selection.[MB1] | An established option when item families, volumes and presentation justify a dedicated pick-and-sort cell.[FAN1] |
| Assembly | Treat as a pilot or roadmap unless the exact subtask has customer-confirmed results. Hyundai’s component-assembly milestone is announced for 2030.[HMG1] | Prefer for low-force, repeatable assembly where human and robot roles can be validated at one workstation.[FAN1][ISO3] | Use to feed the assembly process, not as the assembly mechanism unless a manipulator is engineered and validated.[BMW1] | An established option for stable, precise, high-volume assembly with known variants and controlled interfaces.[FAN1] |
Verified automotive programs
Status checked 23 July 2026
Status is assigned from the customer or OEM source. A manufacturer claim may add detail, but it does not upgrade a roadmap or test into a production result.
| Program | Maturity | Task | What the source establishes |
|---|---|---|---|
| BMW · Figure 02 | Completed pilot | Sheet-metal part positioning for welding | BMW reports more than 30,000 X3 vehicles supported over 10 months, more than 90,000 components and about 1,250 operating hours. BMW describes the program as a completed pilot.[BMW1][BMW2] |
| BMW · Figure 03 | Announced / starting | Component sequencing in assembly logistics | BMW announced the start of the successor project in June 2026. Public acceptance, intervention and scale results were not available at review.[BMW1] |
| BMW · AEON | Pilot / testing | High-voltage battery assembly and component manufacturing | BMW reports laboratory tests and a December 2025 plant test. Its February 2026 release described an April test deployment and a summer pilot phase as planned next steps.[BMW2] |
| Mercedes-Benz · Apollo | Pilot / testing | Intralogistics, parts movement and initial quality checks | Mercedes says Apollo is collecting production-environment data and progressing from teleoperation toward autonomous work at its Digital Factory Campus.[MB1] |
| Hyundai · Atlas | Announced roadmap | Parts sequencing, then component assembly | Hyundai’s published roadmap targets parts sequencing in 2028 and component assembly in 2030. These are future milestones, not current production outcomes.[HMG1] |
| Toyota · KumiPro / ELEY | Research context | Factory manipulation research and production support | Toyota says KumiPro uses HSR-derived technology at Toyota Motor East Japan and describes ELEY as research for future production-site work. This is not evidence of a commercial automotive humanoid deployment.[TOY1] |
Normalized platform evidence
This is not a market ranking. It shows what the reviewed sources establish, keeps robot generations separate and leaves missing evidence visible.
Evidence reviewed 23 July 2026
Swipe to compare all four platforms →
| Buyer field | Figure 03BMW Spartanburg | Apollo / Apollo 2Mercedes-Benz | AtlasHyundai roadmap | AEONBMW Leipzig |
|---|---|---|---|---|
| Automotive evidence | BMW announced a June 2026 start for a sequencing workflow. No outcome metrics were public at review.[BMW1] | Mercedes-Benz is testing Apollo in its production environment; the OEM page does not identify a current robot-generation number.[MB1][APP2] | Hyundai has published 2028 and 2030 automotive milestones. Treat them as a roadmap until deployment results are released.[HMG1] | BMW documents staged testing and a Leipzig pilot for battery assembly and component manufacturing.[BMW2] |
| Relevant task | Pick unsorted components, place them into a sequencing trolley and prepare the trolley for automated onward transport.[BMW1] | Move components or modules to the line and perform initial component quality checks.[MB1] | Material sequencing, assembly and machine tending are named roadmap applications.[HMG1] | Multifunction use with interchangeable hands, grippers or scanning tools in battery and component processes.[BMW2][HEX1] |
| Published payload | Figure publishes a 20 kg payload for Figure 03. Site payload, reach and grasp margins still require task testing.[FIG1] | The original 2023 Apollo announcement published a 55 lb (25 kg) lift. Those values are not carried forward as Apollo 2 facts without current-generation confirmation.[APP1][APP2] | Boston Dynamics publishes 50 kg instant and 30 kg sustained capacity for the product Atlas.[BD1] | Hexagon publishes 15 kg short-term and 8 kg constant carry values for AEON.[HEX1] |
| Energy and availability | Figure publishes a five-hour runtime and 2 kW fast charging. BMW separately describes wireless charging, without publishing a charging-power figure. These are published design claims, not achieved shift uptime.[BMW1][FIG1][FIG2] | The original 2023 Apollo release described four-hour swappable batteries. Apollo 2 says battery swapping is intended to help enable 7×22 operation; that is not evidence of achieved automotive shift uptime.[APP1][APP2] | Boston Dynamics publishes four hours per battery and autonomous battery swapping; site availability remains unverified.[BD1] | Hexagon publishes up to four hours per charge with automatic swapping for continuous-work designs.[HEX1] |
| Workflow integration | BMW pairs the manipulation step with sequencing trolleys, tugger trains or Smart Transport Robots. Public MES and fleet-interface details are limited.[BMW1] | Mercedes describes MO360 data collection, teleoperation and augmented-reality knowledge transfer before autonomous operation.[MB1] | Boston Dynamics says Orbit connects Atlas to MES, WMS and fleet metrics; validate the exact connector and data ownership in contract.[BD1] | BMW describes stepwise integration into its production system; Hexagon presents manipulation, scanning and mission-control capabilities.[BMW2][HEX1] |
| Open buyer evidence | The cited BMW and Figure pages do not disclose intervention rate, p95 cycle time, achieved availability, service SLA, site safety validation or commercial terms for this use case.[BMW1][FIG1] | Exact current generation, achieved autonomy, availability, safety validation, service SLA and commercial terms are not public in the OEM source.[MB1][APP2] | The cited Hyundai and Boston Dynamics pages do not disclose automotive outcomes, intervention data, service SLA or commercial terms. Hyundai’s milestones are future-facing.[HMG1][BD1] | The cited BMW and Hexagon pages did not disclose pilot outcomes, task-level cycle data, site safety validation, service SLA or commercial terms at review.[BMW2][HEX1] |
Procurement hard gates
Score vendors only after the application clears the constraints that protect production continuity, people and quality.
Prove the complete task at required takt, including perception, walking, grasp retries, placement, hand-off and recharge—not a clipped best-case cycle.[BMW1][BMW2][MB1]
Define planned uptime, mean time to recover, restart ownership, safe fallback and the maximum intervention rate the line can absorb.[BMW1][MB1]
Validate the exact part, container, reach, centre of mass, surface, temperature and end effector. A headline payload is not an application rating.[FIG1][APP1][HEX1]
Name the owner and interface for MES, WMS, PLC, fleet, identity, logging, cybersecurity, change control and data retention before the pilot.[BD1][BMW2]
Use the applicable robot and integration requirements, collaborative-operation guidance and machinery risk-assessment method. Require validation and local regulatory review for normal, fault and non-routine work.[ISO1][ISO2][ISO3][ISO4][OSHA1][NIOSH1]
Contract response time, spares, field repair, software support, rollback, model updates, incident handling and responsibility for revalidation after change.[ISO2][OSHA1]
Pilot design
Freeze the baseline, denominator, measurement window and exclusions. A polished demonstration is not an acceptance test.
Successful task completions without human intervention, reported against a fixed denominator.
Median and p95 cycle time against the current manual or automated baseline.
Interventions per 100 cycles, separated into safety, perception, grasp, navigation and integration causes.
Achieved availability during scheduled pilot hours, excluding planned charge and maintenance by explicit agreement.
First-pass quality, damage, mis-pick and sequence-error rates.
Safety events, protective stops and time spent in recovery or manual mode.
Training, integration and operator-support hours required per task or changeover.
A stop rule and rollback path if the robot threatens takt, quality, safety or production continuity.
Total deployment cost
End effectors, safety engineering, simulation and integration can matter as much as hardware. Use your quote and measured operating baseline; there is no responsible generic payback number.
Humanoid ROI and cost calculatorRFP readiness
Record the exact model, generation and configuration in every proposal and acceptance document.
Define one task, route, part family, operating window and baseline before comparing vendors.
List hard pass/fail limits for takt, payload, quality, environment, safety, integration and service.
Require customer-confirmed evidence for deployment claims and manufacturer documents for technical claims.
Separate demo behavior, teleoperation, supervised autonomy and autonomous production operation.
Specify data ownership, remote access, cybersecurity, model updates and validation after software change.
Price the complete deployment and the internal team, not only the robot or monthly fee.
Write acceptance, escalation, stop, exit, asset-removal and rollback terms before the pilot starts.
Buyer answers
Customer-confirmed examples include BMW’s completed Figure 02 sheet-metal handling pilot, newer BMW programs with Figure 03 and AEON, and Mercedes-Benz testing Apollo for intralogistics and initial quality checks. Hyundai’s published Atlas dates are a roadmap, not current production results.
A humanoid is worth testing when the task genuinely needs both mobile access and manipulation across spaces designed for people. Use an AMR for transport-only work, a cobot for a bounded shared station, and fixed automation for stable high-throughput processes when those simpler systems meet the requirement.
Some platforms have completed or entered production-floor pilots, but that is not the same as broad, scaled readiness. Buyers still need task-level evidence for cycle time, availability, intervention rate, safety validation, integration, support and cost.
Measure successful cycles without intervention, median and p95 cycle time, interventions per 100 cycles, achieved availability, first-pass quality, safety stops, recovery time and the engineering hours needed to deploy and change the task.
Use the vendor quote plus end effectors, safety work, integration, site changes, commissioning, training, software, service, spares, insurance, internal labor, downtime and contingency. Compare the resulting total cost with a measured baseline; do not rely on a generic payback promise.
Methodology
RoboZaps uses OEM or customer sources for deployments, manufacturer sources for model claims and standards or regulators for the safety boundary. Public evidence narrows the shortlist; it never replaces the site assessment, pilot or contract.
If a simpler system clears the requirement with less integration, safety and service risk, that is the better shortlist.
Customer and OEM sources set deployment maturity. Manufacturer pages support model-specific claims. Standards and regulator sources define the review boundary; none proves that a specific application is safe.
Customer / OEM · Published 25 June 2026
Reviewed 23 July 2026
Customer / OEM · Published 27 February 2026
Reviewed 23 July 2026
Customer / OEM · Published 18 March 2025
Reviewed 23 July 2026
Customer / OEM · Published 6 January 2026
Reviewed 23 July 2026
Customer / OEM · Published 31 March 2026
Reviewed 23 July 2026
Manufacturer
Reviewed 23 July 2026
Manufacturer · Published 17 July 2025
Reviewed 23 July 2026
Manufacturer · Published 23 August 2023
Reviewed 23 July 2026
Manufacturer
Reviewed 23 July 2026
Manufacturer
Reviewed 23 July 2026
Manufacturer
Reviewed 23 July 2026
Manufacturer
Reviewed 23 July 2026
Standards body · Published February 2025
Reviewed 23 July 2026
Standards body · Published February 2025
Reviewed 23 July 2026
Standards body · Published February 2016; confirmed 2022
Reviewed 23 July 2026
Standards body · Published November 2010
Reviewed 23 July 2026
Regulator
Reviewed 23 July 2026
Government research · Published Updated 25 November 2024
Reviewed 23 July 2026