The First Reported LiDAR Return Is Not Always the Nearest

Identify the returns before interpreting their positions

A return is range information associated with a beam, rather than an identified object. Ouster's Firmware User Manual v3.1.0, dated October 28, 2025, describes Rev7 sensors as detecting up to two returns across the documented profiles. Its sorting explanation identifies the two strongest returns per radial beam, then orders that pair by signal strength or measured range.

Detection, the represented pair, ordering and delivery answer different questions. They are useful distinctions for reading an offer, not a claim about undocumented internal processing stages. Section 10.2.24 says the order also controls the only reported return in single-return UDP profiles. A hardware ability to process two therefore does not mean a recipient receives both.

The explanation here uses that named hardware and firmware scope to interpret supplied data. It does not extend the behavior to other LiDARs or prescribe a preferred order.

References: [O1][O2]

LR-52: the same pair produces three ordered descriptions

LR-52 is an original thought experiment. Stipulate exactly two available returns from one radial beam: A has range 3 m and signal score 100; B has range 8 m and signal score 400. Both are eligible for the illustrated output, with no ties, missing returns or additional filtering. These are invented values, not a sensor capture. The scores are abstract comparable numbers, not material-reflectivity measurements or percentages.

Strongest-first gives B, then A, because 400 exceeds 100. Nearest-first gives A, then B, because 3 m is less than 8 m. Farthest-first gives B, then A.

B happens to be both farther away and stronger in this example. The agreement between strongest-first and farthest-first does not make those rules interchangeable. One ranks signal; the other ranks range.

References: [O1]

One delivered record can mean 8 m or 3 m

Now stipulate a single-return profile that reports the first selected return. With LR-52's strongest-first order, the supplied record represents B at 8 m. With nearest-first, it represents A at 3 m. Both deliveries contain exactly one record.

An offer comparison based only on record count would call these equal. A comparison based on the retained physical return distinguishes them immediately: one contains B and omits A; the other contains A and omits B. The 5 m difference follows from which stipulated return survives into the supplied input. It does not show that either range changed or that either offer is more accurate.

A consumer receiving B alone cannot recover A's stipulated 3 m range merely by knowing that the source hardware can process two returns. The absent return is not encoded by the count of the surviving record.

References: [O2]

Receiving both returns does not make slot one universally nearest

Suppose instead that the supplied input contains both LR-52 returns and preserves their declared order. Strongest-first delivers B at 8 m in slot one and A at 3 m in slot two. Nearest-first delivers the same pair in the opposite positions.

Treating slot one as universally nearest would therefore misread the strongest-first delivery even though no member of the pair was omitted. A count of two cannot reveal that mistaken interpretation. Here the issue is the meaning assigned to a position, rather than the loss of a return.

For a pair that remains fully represented, a range-based comparison can distinguish the nearer member from the stronger member. Once an output retains only one, that same comparison between members is no longer available at the recipient's boundary.

References: [O1][O2]

Nearest describes a member of the represented pair

In the documented Rev7 sorting description, nearest means nearer within the two strongest returns. It does not establish the closest physical surface anywhere in the scene.

Keep a radial beam, a return, a delivered record and an object separate when reading a data offer. In LR-52, A and B label returns; they do not establish two separately identified objects. Likewise, two delivered records do not establish complete physical coverage, and a delivery with no usable return does not establish empty space. The phrase up to two is a limit on the described return population, not a promise of two useful observations for every beam.

These distinctions matter before a count becomes a claim about perception input completeness. No object-detection result, task performance or scene-wide coverage follows from the ordering example.

References: [O1]

Compare offers at the point where the recipient receives data

A useful input description names the hardware revision, firmware version, output profile and declared return order. It also explains whether the supplied input preserves both returns or retains one, and what any later export keeps. If an offer says only one return per beam, the count is specified while the identity rule remains unresolved.

For LR-52, an adequate first-only description lets the reader state either B at 8 m under strongest-first or A at 3 m under nearest-first. An adequate two-return description identifies both the pair and its positions. This is a check on the meaning of the promised input, without choosing a sensor or changing its configuration.

After that meaning is established, the PointCloud2 field-layout guide addresses the next question: how the supplied values occupy fields, point records and rows. Correct byte interpretation and correct return interpretation are both needed; either can be wrong while the record count stays unchanged.

References: [O1][O2]

Scope associated compute-equipment installation separately

If the associated compute equipment also needs physical installation, Leviathan's rack-and-stack service covers an agreed package of equipment mounting, IT cabling, labels and installation records. Use the linked service and project inquiry for that separately scoped physical requirement. Sensor-data semantics and perception acceptance remain with their appointed technical owners.

References: [L1]