A longer range and a faster scan rate can make a sensor look attractive on a comparison sheet. Neither number establishes whether a robot will see a dark pallet leg, localize beside a glass wall, or receive useful measurements before its environment model becomes stale.
This guide focuses on navigation sensing for indoor wheeled autonomous mobile robots. Start with three questions: what must the robot observe, under which conditions, and what should the system do when observations become unreliable?
Separate navigation sensing from protective functions
A navigation sensor supplies measurements for mapping, localization, or obstacle avoidance. A protective function must be designed around the complete detection, decision, and stopping chain. Detecting a person in a point cloud does not, by itself, establish a personnel-protection function.
Laser Class 1 is also a different claim from functional safety. Hokuyo's dated UST-10LX specification lists the laser classification while explicitly identifying the device as unsuitable for use as a safety device. The document is useful as an example of how to read specifications, not as a current purchasing recommendation. Hokuyo specification, March 2015
Nav2's Collision Monitor documentation likewise distinguishes its additional software-based collision monitoring from a safety-certified system. Define navigation acceptance tests separately from the verification of protective functions. Nav2 Collision Monitor
Read range together with its target conditions
The same Hokuyo document gives a 10 m detection range for a white reference target and 4 m for a target with 10% diffuse reflectance. The lesson is the qualification attached to the number: an advertised maximum cannot be applied to every material and angle.
Ask for the target reflectance, incidence angle, ambient-light conditions, measurement uncertainty, and definition of a valid return. Include the actual dark, reflective, and transparent materials in the facility in your test plan. Avoid assuming that an entire material category will always be detected or always be invisible.
Angular spacing matters too. At a distance of 5 m, a 0.25-degree interval corresponds to roughly 22 mm of lateral spacing under a small-angle approximation. This is a geometry illustration, not a guaranteed minimum detectable object size. Beam shape, object position between samples, return strength, and filtering still matter.
Use research to identify failure modes, not to invent guarantees
The TOPGN research paper investigates transparent-obstacle detection using LiDAR intensity information and a layered representation of the point cloud. Its relevance to selection is that the interpretation of returns matters alongside the sensor hardware. A research method evaluated with particular sensors and scenes does not make an arbitrary 2D scanner reliable on glass. Weerakoon et al., TOPGN, arXiv 2024
For a facility with glass partitions, write down the missing-surface problem explicitly. Verify the sensing and processing combination under the intended geometry. If coverage is inadequate, evaluate another sensing modality, a revised installation, or operating constraints with evidence from that configuration. Do not silently turn missing returns into confirmed free space.
Installed coverage is smaller than the nominal field of view
The enclosure, payload, lift mechanism, and wiring can block a sensor that has a wide nominal field of view. Draw the sensor pose on the vehicle outline and inspect coverage with and without a payload, during lifting, and in forward, reverse, and turning motion.
A 2D scanner samples a plane. A low installation can miss an overhanging structure; a higher installation can miss a low obstacle. Chassis pitch, ramps, and uneven floors change the intersection between that plane and the environment. More range cannot recover objects outside the scanned plane.
Distinguish the sensor's specified minimum range from the vehicle's effective near-field blind region. Recessed mounting, body occlusion, and software masks contribute to the latter. When removing self-returns, check that the mask does not erase a nearby region the robot needs to observe.
Scan rate is only one part of data age
Measurements pass through acquisition, transport, driver processing, coordinate transforms, and the navigation environment model. The time until a useful observation reaches a decision point can differ substantially from the scan period.
Consider an illustrative 25 ms acquisition period at 1 m/s: the platform translates 25 mm during acquisition. Rotation adds a changing orientation. This calculation is not a latency measurement or a prediction of range error; it identifies why motion compensation and timestamp conventions may matter.
ROS 2's LaserScan definition assigns the header timestamp to the first ray and provides a time increment between measurements. Check what the device driver actually publishes and how device time is related to system time. Arrival time at the computer should not be mistaken for acquisition time without accounting for the difference. ROS 2 LaserScan definition
Verify both geometry and the use of observations
A visible point cloud is an intermediate result. The sensor pose relative to the chassis must be correct, and the navigation stack must interpret measurements as intended. Changing a bracket or installation angle requires rechecking that relationship. Nav2 transform setup
Nav2's obstacle layer separately configures marking, clearing, input types, and observation ranges. Observe the raw measurements and the costmap together: does a target appear in the right place, disappear correctly when removed, and cause the expected response when data stops arriving? Nav2 obstacle layer
Turn the requirements into repeatable tests
Set target sizes, vehicle speeds, acceptable errors, data-age limits, and repetition counts before testing. The table below is a proposed navigation-integration checklist, not a personnel-protection acceptance standard.
| Test | Conditions to cover | Evidence to retain |
|---|---|---|
| Target visibility | Actual materials, dimensions, distances, and approach angles | Raw scans, missed detections, false detections, and return validity |
| Coverage | Loaded and unloaded geometry, turns, and near-field regions | Installation dimensions, coverage drawings, and remaining blind regions |
| Timing | Acceleration, turning, and representative computer load | Acquisition timestamps, observation age, update intervals, and outliers |
| Coordinate consistency | Known stationary objects with the robot stationary and moving | Raw and transformed locations, calibration records, and configuration version |
| Software behavior | Obstacles added and removed, invalid returns, and interrupted data | Costmap changes, diagnostics, and resulting system behavior |
| Recovery | Restart, contamination, and relevant environmental conditions | Recovery time, maintenance observations, and operating conditions |
Use suitable test objects in a controlled environment, rather than people as collision targets. A useful result states the conditions actually tested and the remaining limitations. That is a stronger basis for comparison than the largest numbers on a data sheet.
Evidence and limits
This is an original engineering synthesis of the linked research and official documentation, reviewed on November 4, 2026. The numerical examples are analytical illustrations; no sensor tests or detection rates are claimed. The 2015 specification is a historical example. Obtain current documentation for the exact product revision before purchasing, and validate the complete vehicle separately.