ArduSimple Offset Error: Fix Lateral Antenna Mounting

Fault Signature and Branch Selection

Cross-track error (XTE) is the perpendicular distance between the guidance reference point and the active AB line. The whole diagnosis turns on how XTE behaves when the machine reverses heading, because each fault class leaves a different fingerprint and only one of them is fixed by touching the antenna.

A pass that reads dead on the line one way and a repeatable 2.4 m out on the return, then dead on again on the next pass in the original direction, is not a correction-quality problem. That pattern is a fixed lateral displacement between the antenna phase centre and the point the software believes it is steering.

Observed pattern	Mechanism	Go to
XTE flips sign on reverse heading, same magnitude both ways	Lateral geometry: antenna off the machine centerline, or offset field mis-set	Check 1
XTE same sign and magnitude regardless of heading	Absolute position shift: base ARP coordinates, datum or epoch mismatch against the reference lines	Check 4
XTE grows by a constant amount every pass, never returns to zero	Guidance width does not match the drilled tramline spacing	Check 2
XTE wanders over minutes, worse on side slopes	Float solution, correction age, or antenna roll with no compensation	Check 4, then Check 3

Check 1: Direction Dependence and the Factor of Two

The term antenna offset here means the signed lateral distance from the machine's guidance centerline to the antenna phase centre, measured perpendicular to travel. The software adds that vector to the reported antenna position to place the machine reference point on the ground.

Leave the offset at zero while the antenna actually sits distance d to one side, and the reported machine centre is displaced by d in machine coordinates — not in field coordinates. Drive north and the reported centre falls d west of truth. Turn round and drive south and the same physical displacement now falls d east of truth. Two adjacent passes driven in opposite directions therefore separate by 2d.

Read the number straight off the field: a repeatable 2.4 m gap or overlap on alternate passes is 1.2 m of lateral displacement. Sign matters. Enter 1.2 m on the wrong side and the passes separate by 4.8 m instead of closing — the doubling is itself confirmation that the fault class is geometric and not a correction fault.

Entering the 1.2 m and driving on is the wrong repair. The software offset restores pass-to-pass agreement while the antenna stays 1.2 m off centre, so implement offsets, headland turn geometry, section control, and any roll compensation added later are all referenced to a point that does not exist on the machine. A roof-centred antenna should never need more than a few centimetres. A metre-scale value in that field means the field is compensating for something else.

Check 2: Configuration Fields Before the Tape Measure

Clear the software before climbing on the roof, in this order.

1. Confirm the vehicle antenna offset field reads 0.00. Working width typed into an offset field is a common entry error; half a working width typed there is the same mistake in a more plausible disguise.
2. Confirm the guidance width equals the drilled spacing exactly — 24.00 m against 24 m tramlines. A width mismatch produces XTE that accumulates by the same amount every pass in the same direction. If the error resets to zero on every second pass, width is not the fault and this branch is closed.
3. Check for double entry. Many systems carry a vehicle antenna offset and a separate tool or hitch lateral offset. Populating both with the same number applies the correction twice; populating them with opposite signs cancels it.
4. Check the antenna fore-aft and height entries too. A fore-aft error does not produce the sign flip, but it does throw line acquisition and turn geometry once autosteer is fitted.

Check 3: Mounting Geometry, Machine Centerline, and Mast Lean

Roof centred is not machine centred. Define the machine centerline from the running gear: the midpoint between the rear tyre or track centres, transferred vertically to roof height with a plumb line. On many cabs the roof panel edge, the handrail, and the beacon boss all sit off that line, and a bracket that hooks a roof rib or bears against a raised panel can carry the antenna a long way out. Metre-scale displacement almost always traces to the bracket, not to a measurement slip.

Mast lean adds the same signature. Lateral error from tilt is h × sin(θ), where h is antenna height above the reference plane and θ the tilt from vertical. On a lightbar-only install with no IMU, nothing removes that term: a leaning mast injects a constant lateral bias that reverses with heading exactly like a mounting offset, and machine roll on side slopes stacks a variable bias on top. Level the mast on two axes with the machine on flat ground, and keep the antenna as low as an unobstructed sky view allows.

The reversal test isolates geometry from everything else. Park with a machine reference mark over a driven nail, log the antenna position, then drive out, turn, and return with the same mark over the same nail on the opposite heading. The two logged positions differ by twice the lateral offset. This test needs no field, no AB line, and no assumptions about the operator's steering.

Check 4: Fix Quality, Base Coordinates, and Steering Bias

RTK fixed means the carrier-phase integer ambiguities are resolved; float means the solution is still estimating them as real numbers. In NMEA GGA, quality indicator 4 is fixed and 5 is float. Float will show decimetre agreement for a few minutes and then drift, which makes it useless as a basis for setting geometry and useless as evidence of 10 cm pass-to-pass performance. Take every geometry measurement at quality 4 with differential age under a few seconds.

When the receiver will not hold a fix on an NTRIP caster feed, work through the causes in this order: correction age climbing (caster outage or mobile link dropout), the RTCM 3 message set the mountpoint actually sends (1005 or 1006 for the base antenna reference point, MSM observation messages such as 1074/1084/1094/1124 for the constellations you track, 1230 if GLONASS is in the solution), baseline length — ionospheric decorrelation over tens of kilometres blocks ambiguity resolution — and multipath off the cab roof, exhaust stack, and beacon.

Base coordinates decide absolute agreement, never the sign flip. A base broadcasting an antenna reference point (ARP) that is a metre out shifts every pass by the same metre in the same direction, whichever way you drive. When your lines have to match tramlines drilled from a different RTK base, that shift plus any datum or epoch difference between the two base positions is the real disagreement — and it does not alternate.

Steering bias belongs on a separate branch. With lightbar-only guidance there is no wheel angle sensor (WAS) in the loop, so a WAS zero error cannot produce the offset; the operator closes the loop by hand. Once autosteer is added, a WAS zeroed off centre holds a standing steer angle and the controller settles on a line displaced to one side, reversing with heading like a mounting offset. Separate them by steering manually along the AB line and watching XTE: a geometry error survives manual steering, a WAS zero error does not.

Resolving Branch: Remount and Verification

1. Set the antenna offset field to 0.00 and the working width to the true implement width before taking any measurement.
2. Mark the machine centerline from the midpoint between rear tyre or track centres and transfer it to roof height with a plumb line.
3. Fabricate the mount so the antenna phase centre — the marked centre of the housing, not a corner of the bracket — sits over that line. Reference the plate to the centerline, not to a roof rib, handrail, or panel seam.
4. Level the mast in both axes on flat ground; keep any ground plane horizontal.
5. Route the antenna coax clear of the exhaust and beacon wiring, and away from any transmit antenna.
6. Power up and wait for RTK fixed before evaluating anything.

Check 1: tape from the left tyre or track centre to the antenna centre, then from the right. Expect the two readings equal within 20 mm.

Check 2: repeat the reversal test over a driven nail. Expect the two logged positions within a few centimetres; any separation is twice the residual lateral offset.

Check 3: read GGA quality indicator throughout. Expect 4A5 invalidates the test — stop and fix the correction path first.

Check 4: drive the known 24 m tramlines in both directions with the offset field at 0.00. Expect the same XTE sign and magnitude on both headings, inside the noise of the fix. A residual sign flip of x means x/2 of lateral offset is still in the mount — return to Check 1 rather than typing it into the offset field.

Check 5: hold the fix for an hour of work and log XTE alongside correction age. Expect pass-to-pass agreement inside 10 cm, with excursions only where correction age spikes; excursions that correlate with heading rather than with age mean geometry is still wrong.

Frequently Asked Questions