The direct answer
Does a large public specification table mean robot-component performance is independently verified? No. In RCI catalog v0.4.0, 525 of 544 claims—96.5%—are supported by manufacturer-controlled sources. One claim, covering observed RealSense D455 depth-error behavior, has independent study evidence. Three claims contain unresolved disagreements between official sources, one is an explicit RCI derivation and fourteen are published as unknown.
Does 83% field coverage mean the products are directly comparable? No. RCI found values in 377 of 452 selected comparison cells, but filled cells still mix continuous, rated, peak and stall torque; ideal and instrumented range; sparse and dense TOPS; module and system power; and cell-level versus pack-level limits. Field presence measures documentation availability, not equivalence.
These results describe the 75 exact records, six comparison definitions and cited sources in dataset v0.4.0 as checked on 2026-08-10. They are not a market census, brand score or estimate of every product a manufacturer sells.
What RCI counted
The audit spans 25 manufacturers and five robot-hardware layers: motion and actuation; perception and sensing; compute and control; power and energy; and interfaces and integration. A claim is one source-linked field with its original value, unit, condition, evidence state, source location, verification date and editorial limitation.
For the comparison audit, RCI fixed six fields for each of six evidence tables. A cell counted as filled when the selected component record contained a claim using that field. The method does not award extra credit for more impressive numbers, cleaner marketing pages or a better-known brand.
| Hardware layer | Components | Claims | U | C | V |
|---|---|---|---|---|---|
| Motion & actuation | 18 | 130 | 3 | 0 | 0 |
| Perception & sensing | 21 | 146 | 2 | 1 | 1 |
| Compute & control | 17 | 123 | 4 | 1 | 0 |
| Power & energy | 11 | 79 | 2 | 0 | 0 |
| Interfaces & integration | 8 | 66 | 3 | 1 | 0 |
Manufacturer evidence dominates every hardware layer
| State | Claims | Share | What it means here |
|---|---|---|---|
| M · Manufacturer claimed | 525 | 96.5% | An official page, manual or datasheet supports the reported value. |
| V · Independently verified | 1 | 0.2% | An independent study directly tested the claim or behavior under a stated setup. |
| D · RCI derived | 1 | 0.2% | RCI calculated the value from cited inputs and publishes the formula boundary. |
| C · Conflicting | 3 | 0.6% | Current-facing official sources disagree and RCI preserves both values. |
| U · Unknown | 14 | 2.6% | The selected official source does not publish the material field. |
The single V record does not convert the D455’s whole specification sheet into independently verified data. It supports one bounded finding from a published study: depth noise increased with distance and accuracy degraded at high surface inclinations under the paper’s setup. The product page’s full operating range and accuracy claims remain manufacturer evidence.
That narrowness is deliberate. RCI assigns evidence state per claim, not per company, document or product. A peer-reviewed paper can verify one behavior without validating every headline specification, just as an official datasheet can remain the strongest source for model identity while still representing a manufacturer claim.
The fullest table is not necessarily the safest comparison
| Evidence table | Records | Filled cells | Coverage | Largest selected gap |
|---|---|---|---|---|
| Integrated joint actuators | 18 | 96 / 108 | 89% | Reduction ratio: 6 records |
| Stereo and depth cameras | 11 | 60 / 66 | 91% | Synchronization: 3 records |
| 3D LiDAR | 9 | 43 / 54 | 80% | Channels and power: 4 records each |
| Main-compute platforms | 17 | 71 / 102 | 70% | Software: 8; camera I/O and power: 7 each |
| Battery packs and cells | 11 | 55 / 66 | 83% | Certification: 5 records |
| Host-to-CAN interfaces | 8 | 52 / 56 | 93% | Software: 3 records |
Host-to-CAN interfaces reach 93% selected-field coverage, but the table still mixes USB, Ethernet and PCIe host boundaries and both classical CAN and CAN FD. Depth cameras reach 91%, yet range language spans ideal range, operating range and lens-specific depth range. High coverage makes conditions more visible; it does not erase them.
Compute has the lowest coverage at 70%. Eight of seventeen records lack a claim in the selected software field, seven lack a carrier- or system-level camera-I/O claim and seven lack a directly usable power-envelope claim. These absences matter because a compute module can advertise high accelerator throughput while leaving the robot designer to resolve carrier routing, supported software images and peripheral power.
Three official-source conflicts are revision-control signals
The catalog preserves three current-facing disagreements instead of selecting the tidier value:
- Kvaser USBcan Pro 2xHS v2: the product page states a 20 kbit/s minimum classical-CAN bitrate while the cited user guide states 50 kbit/s.
- Hesai QT128: the current product page states 1,152,000 points/s in single-return mode while the Q03 manual dated 2024-11-20 states 864,000 points/s.
- Advantech MIC-713-OX: the datasheet updated 2026-07-30 states up to 157 TOPS with Super mode while the current product page still states up to 100 TOPS.
None of these conflicts proves that either value is fabricated. They indicate that product pages, manuals, firmware modes and hardware revisions can move at different speeds. The procurement response is to request the exact ordering code, hardware revision, firmware or software mode and a dated supporting document.
What a buyer should ask after reading a populated table
- Identity: Which exact orderable SKU, lens, memory size, connector, bus variant and hardware revision does the row represent?
- Condition: At what voltage, current, temperature, cooling setup, target reflectivity, distance, precision and software mode was the value defined?
- Duration: Is the rating continuous, rated, peak, stall, startup or a protection threshold—and for how long?
- Integration: Does the value describe a bare cell, module or complete system, and which carrier, cable, charger, driver or deserializer is required?
- Revision: What document revision and product revision will ship, and is there a dated replacement or end-of-support path?
- Independent evidence: Is there a reproducible test, certification record or field dataset that measures the decision-critical behavior rather than repeating the vendor table?
A product with unknown or conflicting fields is not automatically inferior. It is a product with unresolved evidence work. Making that work visible is more useful than replacing it with a composite score.
Download and reproduce the brief
Download the immutable v0.4.0 JSON catalog or the flattened claim CSV. The methodology defines the evidence states and publication gate; the change log records material revisions.
Suggested citation: Robot Component Index, “75 Robot Components Audited: 544 Claims, One Independently Verified Result,” RCI Data Brief 001, dataset v0.4.0, 2026-08-10.