Does Your AMR Need Dual-Antenna GNSS to Know Its Heading at Standstill?

The Position Is Correct. The First Turn Is Wrong.

An outdoor autonomous mobile robot wakes up beside a loading bay. Its position marker sits inside the correct parking space. The satellite receiver reports a high-quality position. The dispatch system releases the next mission. Yet the first steering command points across the bay rather than toward the exit.

This is an illustrative failure scenario, not a reported incident. It exposes a question that a position-accuracy specification cannot answer: does the robot know which way its body is facing before it moves?

For a vehicle that always starts in a surveyed dock and follows a clear forward departure, the answer may be straightforward. For a robot that restarts in an open yard, reverses out of a transfer station, or moves sideways under a load, the answer needs more evidence. A correct point on a map does not establish the orientation of the vehicle, its forks, or its payload.

The practical purpose of dual antenna GNSS is to add an orientation reference that does not depend on the robot first travelling along its forward axis. Whether that capability earns its cost depends on the actual startup, parking, reversing and transition tasks. This guide turns those tasks into questions a buyer can put into a supplier demonstration and a purchase specification.

Why Stationary Heading Has Become a Timely Buying Question

Several recent announcements make this a relevant procurement discussion in late 2026. On September 3, Xsens announced a dual-antenna RTK GNSS/INS solution in Avior and Sirius configurations, explicitly addressing true heading at standstill and low speed. Its announcement describes integration with the u-blox ZED-X20D receiver and says ordering and availability will follow the introduction. That is evidence of a product direction, not proof that every configuration is already shipping or qualified for an AMR application. [1]

The September 23 u-blox ZED-X20D presentation likewise focuses on motion-independent heading. One especially useful detail is its qualification of a static heading result: the stated statistic is relative to an empirical mean and represents consistency, rather than an independent measurement of absolute angular accuracy. Buyers should preserve that distinction when comparing a presentation chart with a site acceptance requirement. [2]

On September 28, Septentrio announced additional partner products and open-source reference designs around its mosaic-G5 family, including integration paths for robotics development platforms. This is a broader signal that high-precision satellite positioning is becoming easier to integrate. It does not establish that a particular board provides dual-antenna heading, nor that an assembled robot can operate through satellite obstruction. [3]

These developments justify asking a sharper question about outdoor AMR navigation. Instead of requesting another impressive outdoor driving video, ask the supplier to begin with the robot powered down, physically oriented in a different direction, and prevented from moving until it has established an acceptable heading. The resulting evidence is much closer to the operational problem than a long straight run after the estimator has already settled.

Separate the Three Arrows on the Navigation Screen

GNSS heading vs course over ground

The distinction behind GNSS heading vs course over ground is simple but consequential. Course over ground describes the direction of translation. Body heading describes orientation. A dual-antenna receiver measures the direction of the baseline between its antennas; the system must relate that direction to the vehicle body. NovAtel's azimuth documentation separates these quantities and explains that course over ground is undefined at zero velocity. [4]

Imagine a robot reversing straight out of a station. Its direction of travel is opposite to the direction its front faces. Now imagine an omnidirectional base sliding sideways. The travel vector and the forward body axis differ again. A controller that treats the travel arrow as the body arrow can receive plausible numbers with the wrong physical meaning.

A third arrow may represent a planner's desired direction. That command is neither a measured heading nor proof that the chassis followed it. During a demonstration, ask for all three quantities to be labelled separately: measured body orientation, observed travel direction, and commanded motion. If the dashboard shows only one rotating triangle, request the underlying signal definitions before drawing conclusions.

This is why a complaint about GPS heading at low speed should not immediately trigger a receiver replacement. The first investigation is whether the application is reading a course field, an antenna-baseline heading field, a fused inertial heading, or a default value held since startup. Changing hardware while preserving the wrong interface interpretation can preserve the failure.

Position quality and orientation quality need separate release conditions

Searches for RTK GPS heading often mix two requirements: accurate global position and accurate orientation. GPS is one satellite constellation within GNSS; RTK describes a positioning technique, not a universal heading guarantee. A position quality flag should therefore not be used as the sole permission to start a heading-sensitive manoeuvre.

Request a supplier-defined readiness record with separate fields for position validity, heading validity, reference frame, measurement age, and the current alignment state. The application should know whether it is observing a new valid estimate, a prediction carried forward from an earlier estimate, or an invalid output retained for display. A stationary vehicle needs a meaningful orientation reference even when its reported speed is exactly zero.

The implementation should also distinguish unavailable heading from an angle of zero. Zero can be a real orientation. Reusing it as an undocumented failure code creates an avoidable ambiguity at exactly the moment when the robot is deciding whether to move.

A concrete example appears in the documented u-blox X20 HDG 2.00 interface. Its UBX-NAV-DAHEADING message includes a baseline heading from antenna 1 toward antenna 2. The heading is valid only when the relPosHeadingValid flag is set; accHeading reports the associated accuracy figure. A configured heading offset can express the vehicle's forward direction, but it does not rotate the other reported baseline-vector components. These are specific interface rules, not field names shared by every receiver. [9]

That example suggests a useful procurement check: have the supplier trace one actual receiver message through the driver and estimator to the controller's readiness decision. Confirm that validity, units and frame meaning survive the complete path. A correctly wired sensor can still be misused by an integration that forwards the angle but discards its status.

Build the Business Requirement Around the First Metre

The most revealing specification for robot heading at standstill starts with the first movement after each operational pause. List where the robot can be when it boots, which direction it may face, what load it carries, and what manoeuvre follows. A restart inside a wide outdoor staging area has different consequences from a restart beside a gate post with an overhanging pallet.

For each starting location, determine whether a preliminary alignment movement is possible and acceptable. If the robot has to move to establish its orientation, that movement itself needs a justified operating condition. A commissioning technician's informal instruction to drive forward until the arrow looks right is not a production startup procedure.

A simple geometry example explains why angular error deserves its own allowance. Suppose a point of interest lies two metres ahead of the robot's orientation reference. A one-degree yaw error corresponds to about 35 millimetres of lateral displacement at that point, before adding position error, mechanical deflection or tracking error:

lateral displacement = distance × sin(yaw error)
2.0 m × sin(1°) ≈ 0.0349 m

This is an illustrative geometric calculation, not a recommended acceptance limit. Use the actual distance to a fork tip, load corner, or relevant control point. A longer assembly can consume more clearance with the same angular error, while a compact vehicle in an open area may tolerate more. The calculation gives procurement and engineering a common way to discuss why a heading requirement exists.

Then translate the requirement into observable behaviour. The robot may need to establish heading before releasing a mission, hold its departure while orientation remains unresolved, or use a separately validated local reference for a particular station. Avoid specifying only an average angle error over a complete route. A low route average can hide an unacceptable first command after a restart.

Also define the reference for acceptance. A dashboard agreeing with its own stored trajectory is not an independent angular measurement. Agree how the test will establish the real body axis and how uncertainty in that reference will be reported. Otherwise the project may mistake repeatability around a biased angle for accuracy.

Choose an Orientation Architecture Before Choosing a Module

The decision between single vs dual antenna GNSS is not a contest in which two always wins. The correct comparison is between complete sources of usable orientation, including their startup conditions and failure behaviour. A single-antenna installation can still provide a credible heading through an appropriately designed inertial or map-based system.

NovAtel's startup documentation illustrates the range of possibilities: static alignment where the inertial hardware supports it, alignment using motion, dual-antenna input, a retained alignment state, and an externally supplied initial azimuth. These are product-specific options with conditions; they should not be generalized into a promise that every low-cost IMU finds north while stationary. [5]

Use this table to eliminate unsuitable architectures before comparing price. A vehicle with a very short available mounting span might be a poor match for a quoted dual-antenna specification measured on a much wider platform. A yard robot that routinely parks under a metal canopy might need a strong local orientation source even if its outdoor satellite performance is excellent.

Conversely, a slow robot working in open space may gain operational simplicity from an absolute heading that does not require an alignment drive. That benefit can be valuable even when another architecture is technically possible. The commercial question is whether it reduces startup delay, operator intervention, or restrictions on where the robot can be restarted.

When comparing GNSS INS for robots, ask which output the controller actually consumes. A raw dual-antenna heading and the fused inertial orientation are different signals. NovAtel explicitly documents separate outputs and the configuration needed to connect antenna geometry to the inertial frame. A purchase specification should identify the delivered interface and its status reporting, rather than merely naming the sensors inside the enclosure. [6]

The Antenna Installation Can Decide the Result

Baseline length is a design input, not a catalogue footnote

The GNSS antenna baseline is the physical separation used to determine the relative antenna direction. NovAtel's installation guidance links heading performance to that separation and calls for stable mounting and a clear sky view. Its numerical examples apply to the documented products and configurations; they are not universal minimum distances for every dual-antenna receiver. [7]

For yaw, consider the baseline's horizontal projection, not simply the distance measured between antennas. Increasing their height difference does not provide the same heading geometry as increasing their horizontal separation. Review the installed geometry with the supplier, including the vehicle's permitted pitch and roll. [4]

Ask the supplier to mark the proposed antenna positions on the production robot drawing. Include the maximum load envelope, lifting motion, service panels, protective housings and any removable accessories. A layout that works on an empty demonstration base may be unsuitable once a tall metal load sits between an antenna and the sky.

The mounting span should remain mechanically defined throughout normal use. If one antenna sits on a panel that technicians remove, the service procedure must restore the intended geometry. If an antenna is mounted on a moving assembly, the integration must explicitly account for that motion. Treat an unmeasured bracket replacement as a navigation configuration change, not merely a cosmetic repair.

Absolute accuracy is different from a steady arrow

A convincing demonstration of dual antenna GNSS heading accuracy should report both variation and offset from an independent reference. An arrow can remain very steady while pointing consistently away from the real body axis. Ask whether a quoted RMS value describes error against truth, variation around an average, or an estimated uncertainty produced by the receiver itself.

Also request the antenna type, separation, mounting arrangement, firmware, output rate and environmental conditions associated with the result. Avoid translating a roof test into a promise for a compact robot travelling beside a warehouse wall. The supplier should demonstrate why the published result is relevant to the installed configuration or provide a separate measurement.

The acceptance record should preserve any exclusions. If a test omitted periods when heading was invalid, report the excluded duration alongside the accuracy statistic. Otherwise a system that provides a good angle only intermittently may appear equivalent to one that remains available throughout the operational task.

Agree what the baseline points toward

Label the primary antenna, secondary antenna, body-forward axis and sensor axes on the same drawing. Identify the angular transformation from the antenna baseline to the vehicle and the physical offsets used by the estimator. Do not assume that swapping antenna cables is harmless because the satellite position still looks reasonable.

Verify coordinate conventions at the software boundary. A navigation component may expect a different zero direction, rotation sign or angle unit from the receiver output. Confirm the transformation with several physically distinct orientations, including values near the angle wrap boundary. A check performed only when the vehicle points roughly north can miss a configuration error that appears after a turn.

The official Nav2 GPS tutorial provides a useful integration example: it discusses the yaw offset needed for the expected convention and the need for a consistent datum when GPS positions are combined with an existing map. Those are configuration responsibilities, not capabilities automatically supplied by buying an accurate receiver. [8]

Take the Demonstration Back Through the Warehouse Door

Motion-independent heading is not obstruction-independent heading. Before approving a dual-antenna installation, identify the portions of the route where either antenna loses the observations needed by the receiver. A warehouse doorway is especially useful as a test location because it connects two different operating environments and often includes a pause, a turn or a change in traffic permissions.

Make the supplier explain which source establishes orientation in the yard, which source carries it through the doorway, and which source establishes it inside. The answer may include inertial propagation, wheel odometry, visual information, or map-based localization. The important point is that the handover must exist as a defined design, with observable status, rather than as an assumption that one good outdoor sensor covers the entire route.

Use the existing guide to AMR mapping and localization for the distinction between a map and a live pose estimate. In this application, additionally ask how the indoor map orientation is related to the outdoor reference. Two internally consistent systems can still disagree at the doorway if their coordinate relationship is wrong.

Test a restart under the canopy as a separate condition from a robot that drives under it after aligning outdoors. The latter begins with an established orientation and may carry that estimate for a limited period. The former may lack the same starting information. Combining both cases into a single indoor-outdoor success rate hides a meaningful operating restriction.

Likewise, test satellite-observation loss separately from correction-service loss. The mechanisms are different, and the product may react differently to each. Ask for the actual receiver and estimator behaviour rather than assuming every loss of external data produces the same heading or position state.

The release decision after observations return also needs definition. A newly accepted heading can disagree with the estimate used during the interruption. The application must handle that disagreement in a way consistent with its validated motion envelope. Our AMR localization confidence guide covers the broader confidence policy; here, retain separate evidence for angular disagreement, angular correction and the first subsequent motion.

A Demonstration Script That Exposes the Expensive Failures

The following sequence is an editorial proposal for supplier evaluation, not a standards-mandated test. Engineering and safety teams should adapt it to the vehicle, site and permitted operating modes. Define pass criteria before the demonstration and record unsuccessful attempts as well as successful ones.

Start without inherited orientation

Place the robot in a controlled test area and establish an independent body-heading reference. Run genuinely different initialization conditions agreed with the supplier: a normal reboot, a restart after a longer shutdown, and any approved maintenance reset. Record whether retained state or a known docking orientation was available. A demonstration that silently benefits from yesterday's heading does not answer a requirement for an unknown startup orientation.

Repeat with the robot facing different directions and with relevant loads installed. Measure time to valid position, time to valid heading, time to application readiness, and the reason for any motion inhibition. These clocks answer different questions. A fast receiver output is not useful if the controller cannot establish which frame it represents.

Stationary initialization does not mean zero-time initialization. The receiver still needs suitable observations and a valid solution. Record the waiting interval instead of starting the demonstration clock after the angle has settled.

Remain still, then move in the inconvenient direction

Hold the robot stationary long enough to represent its actual waiting task. Observe the output, the reported uncertainty and the validity state. Then perform the permitted reverse, turning or sideways manoeuvre. Compare body orientation with travel direction and ensure the application continues to interpret them correctly.

Include starts near the slowest operating speed rather than testing only a convenient cruise speed. Examine the first commanded turn and the resulting physical motion. If the supplier attributes an error to a minimum-speed requirement, capture that requirement explicitly and check whether operations can meet it at every relevant station.

Change the installation conditions deliberately

Repeat representative starts with the maximum permitted payload and with the robot positioned near the structures that matter on the route. Use approved methods to evaluate antenna obstruction and sensor faults in a controlled setting. Monitor whether the system detects the condition, whether it changes its validity reporting, and whether the application responds as specified.

Evaluate the service process as well. Have the responsible team perform an approved antenna or bracket replacement, then verify the configuration and heading again. This reveals whether the installed accuracy depends on undocumented workshop knowledge that will disappear after the first maintenance visit.

Measure recovery without concealing the interruption

For a defined loss and restoration of observations, record the full interval: initial degradation, detection, application response, restored measurements, accepted orientation and release to motion. Include the magnitude of the correction and any requirement for operator action. A log that starts only after heading becomes valid cannot establish how the system managed the preceding uncertainty.

Preserve synchronized receiver, estimator and controller records, together with the relevant independent measurements. The record should show firmware and configuration versions, antenna geometry, test conditions and the exact application criteria. A reviewer should be able to determine why movement was permitted without needing the original demonstrator to narrate the video.

Convert the Findings into a Purchase Decision

A useful request for quotation asks for a capability statement tied to the intended robot. Specify the startup locations, permitted orientations, available mounting space, load configurations, shortest alignment opportunity and required first movement. Then ask the supplier to declare the conditions under which the proposed system meets those needs.

Separate mandatory functions from optional convenience. If stationary heading is essential for restarting beside a narrow transfer point, make it an acceptance requirement. If it merely shortens a startup sequence in an open staging area, evaluate its economic benefit. This prevents an attractive feature from becoming a vague premium that nobody can connect to throughput or intervention time.

Compare installed cost rather than the receiver price alone. Include antennas, brackets, cables, integration, configuration control, any positioning-correction subscription, service training and the acceptance work. A correction service may be required by the chosen global positioning architecture; do not automatically assume that the same service is required by every relative-heading implementation. Obtain a product-specific explanation.

The value calculation should use measured operational quantities. Record startup-related interventions, the time spent resolving them, any restricted parking locations and any interrupted handoffs. Compare the demonstrated change against the full cost of ownership. Do not insert a claimed productivity percentage simply because the new module produces a more stable display.

Keep the safety claim separate from the availability claim. A general-purpose GNSS or inertial output does not by itself establish a safety-rated heading function, safe protective-field selection or safe autonomous restart. If orientation contributes to a safety-related decision, the evidence must address the complete function and its failure response. The related AMR safety sensing evidence audit explains how to assess that boundary.

The preferred architecture is the one that establishes the required orientation under the permitted startup conditions, exposes its limitations, and remains maintainable on the production vehicle. Dual-antenna technology is particularly useful when stationary outdoor heading closes a real operational gap. Its strongest justification is a verified first move, not the presence of a second antenna.