FIELD ENGINEERING UWB LOCALIZATION / 01
See the position.
Understand the uncertainty.
A virtual greenhouse. A mobile tag. An observable experiment.
Explore how anchor geometry and biased ranges change the position you trust—then compare estimators on the same measurements.
Localization observatory
SIMULATIONASK A BETTER ENGINEERING QUESTION
Configure the experiment
Change one factor. Apply it. Observe what changes.
Measurement model
10 HzIndependent Gaussian range noise plus a constant positive bias on one selected anchor. These are controlled assumptions.
Surveyed anchor positions
METRES| Anchor | X | Y | Z | Use |
|---|
Select an anchor, then click its new position in plan view.
Coordinates are assumed exact. The solver uses 3D anchor-to-tag distance while estimating X and Y.
Prescribed inspection route
18 × 10.8 m0 waypointsClick the plan view to add a point.
Prescribed tag path; no robot dynamics or crop-avoidance model.
The rover visualizes the experiment. It does not navigate from the UWB estimate.
FROM MEASUREMENT TO POSITION
Inspect the evidence
Residuals describe fit. Position error describes accuracy.
Current range observations
SELECTED SAMPLE| Anchor | Measured m | NLS residual m | Huber residual m | Huber weight |
|---|
A small residual does not prove that a range—or a position—is unbiased.
Position error along the route
METRES / SECONDSBoth fits use identical range observations. The reference position is used for scoring only.
04 / REPRODUCIBLE COMPARISON
One route is an illustration.
Twenty seeds reveal the pattern.
Compare ordinary and robust fitting across twenty paired full-route simulations, using the applied anchor survey and range model.
Use error statistics, availability, and nominal ellipse coverage together. Robust fitting is not guaranteed to improve every error metric.
| Estimator | RMSE m | P95 m | Maximum m | Valid % | Nominal ellipse coverage % |
|---|
Pooled error statistics use each estimator’s valid solves; compare availability as well. NLS ellipse coverage is a nominal LOS diagnostic, not calibrated under bias. Coverage is not a 95% calibration test when σ = 0.
TAKE THE EVIDENCE WITH YOU
Ready for a technical discussion.
Export the measurements, assumptions and figures behind the demonstration.
Simulated known-height planar UWB localization with surveyed three-dimensional anchor positions. All reported performance is generated by the specified measurement model and is not a field measurement.
Model, interpretation & provenanceWhat the experiment represents—and where its evidence ends.+
OBSERVATION MODEL
3D ranges. A known-height position.
The tag height is known and fixed. The estimator recovers only horizontal X and Y from anchor-to-tag slant ranges:
Range noise is independent Gaussian noise with standard deviation σ. A user-selected anchor can have an additional constant positive range bias. Anchor coordinates are treated as exactly surveyed.
Both estimators receive only anchor coordinates, ranges, σ, and the known tag height. They do not receive the reference trajectory or true position.
ESTIMATION & GEOMETRY
Fit is not the same as confidence.
Nonlinear least squares minimizes squared range residuals. The robust fit uses a Huber loss with a transition at 1.5σ (and a small numerical floor when σ = 0). Both methods use the same measurement-derived candidate starting points; neither is temporally filtered.
The ellipse is centered at the NLS estimate and is a local, nominal unbiased-Gaussian LOS approximation. Selecting Huber does not turn it into a Huber confidence region. It is not calibrated uncertainty in the presence of bias.
Near-collinear anchors can produce mirror ambiguity. A local ellipse cannot reveal every alternative position solution.
EXPERIMENTAL SCOPE
An inspectable simulation.
The reference trajectory is prescribed and used only to generate observations and score estimated positions. Heading follows the route; UWB does not estimate the displayed robot heading.
The model does not simulate RF propagation, multipath, clock offsets, packet loss, antenna offsets, survey errors, sensor calibration, vehicle dynamics, or obstacle avoidance. The greenhouse and rover are explanatory 3D geometry, not measured perception. The diagonal and custom paths use an empty test floor so the prescribed path can be examined without crossing the illustrated benches. Range links show ideal geometric connections, not noisy range radii. The geometry map shows nominal horizontal RMS from the local covariance, saturating at 0.50 m; invalid cells are gray.
Use the experiment to examine sensitivity and design a test. Deployment performance requires measurements from the intended hardware and site.
Technical reading
These sources provide context for UWB ranging and localization geometry. The demonstration is an independent illustrative model, not a reproduction of published field results.
Software acknowledgements
ADAPTIVE AGROTECH — UWB LOCALIZATION STUDIO 3D Third-party notices Three.js r128 and OrbitControls Source: https://github.com/mrdoob/three.js License: MIT MIT License Copyright 2010-2021 Three.js Authors Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. Original Adaptive AgroTech branding is preserved from the original supplied logo. All field and robot geometry is procedural. No remote dependencies are used.