⚡ electrical engineering

Bridging the Indoor-Outdoor Gap: Cross-Technology Ranging for Seamless Robot Navigation

This paper presents preliminary findings from the publicly available HYMN dataset, which time-synchronizes raw GNSS, UWB, WiFi, and BLE measurements against millimeter-level ground truth to demonstrate that these technologies are complementary and to characterize their performance degradation at indoor-outdoor boundaries.

Original authors: Paul Schwarzbach

Published 2026-04-29

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Paul Schwarzbach

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a robot that needs to deliver a package. It starts outside in the sunny parking lot, then walks through a large open door, and finally enters a crowded warehouse.

The problem is that the robot's "GPS" (its sense of where it is) works perfectly outside but goes haywire the moment it steps inside. Conversely, the robot's "indoor map" (using local signals like Wi-Fi) works great inside but is useless outside. The worst part is the doorway itself: that's where both systems fail at the same time.

This paper is like a detailed report card for four different "rulers" (technologies) that robots use to measure distance. The author, Paul Schwarzbach, gathered data from a real industrial site in Germany to see how these rulers behave when a robot moves from outside to inside.

Here is the breakdown using simple analogies:

1. The Four "Rulers"

The study tested four different ways to measure distance, each with its own personality:

GNSS (Satellites): Think of this as looking at the stars to find your way.
- Outside: It sees 20–30 "stars" (satellites) and knows exactly where it is.
- Inside: The roof blocks the view. It sees zero stars.
- The Doorway: It sees only a few stars peeking through the door, making the reading shaky and unreliable.
UWB (Ultra-Wideband): Think of this as a high-tech walkie-talkie that bounces signals off specific anchors (like lighthouses) placed around the building.
- Inside: It's very accurate, like a laser measure.
- Outside: It can only see the lighthouses near the door, so the view is limited.
- The Doorway: It starts working well, but the signal gets a bit "noisy" as it passes through the wall.
WiFi & BLE (Bluetooth): These are like listening to the hum of nearby routers or beacons.
- They are available almost everywhere (inside and out), but they are often "biased." Imagine a ruler that is always 5 meters too long. They also get confused easily by walls and reflections (multipath), making their readings "heavy-tailed" (meaning they sometimes spit out wildly wrong numbers).

2. The "Doorway" Problem

The paper's biggest discovery is about the transition zone (the doorway).

Usually, we think: "When the GPS fails, the indoor system takes over." But the data shows that at the exact moment a robot crosses the threshold, both systems are struggling.

The GPS is losing satellites.
The indoor anchors are only visible from a bad angle.
The signals are bouncing off the doorframe and walls.

It's like trying to drive a car where the GPS signal is fading out just as the road signs are becoming blurry. If the robot's computer doesn't know this is a "danger zone," it will crash or get lost.

3. The "Residual" (The Mistake)

The authors measured the "residual," which is simply the error in the measurement (how far off the ruler is from the truth).

UWB is usually very honest, but when it lies (due to walls), it lies big (heavy-tailed errors).
WiFi and Bluetooth are consistently "optimistic" or "pessimistic" (they have a constant bias, like always adding 5 meters to the distance).
GPS is consistent but has a large, steady error when it's struggling near the building.

4. The Solution: A Smart "Traffic Controller"

The paper suggests that robot software shouldn't just trust all rulers equally. Instead, it needs to be zone-aware:

When outside, trust the GPS.
When inside, trust the UWB.
Crucially: When at the door, the robot needs to be extra careful. It should expect errors, ignore the "weird" outliers, and perhaps rely on a mix of all four systems while knowing that the data is "noisy."

Summary

This paper doesn't invent a new robot or a new sensor. Instead, it provides a map of the "blind spots." It tells us that the transition between outside and inside is a chaotic place where all technologies degrade simultaneously. To navigate this, robots need software that understands the specific "personality" and "bad habits" of each technology, rather than treating them all as perfect.

The data used for this study (called the HYMN dataset) is now public, allowing other researchers to build better "traffic controllers" for robots that need to move seamlessly between the world outside and the world inside.

1. Problem Statement

Mobile robots operating in logistics, inspection, and search-and-rescue frequently transition between outdoor and indoor environments. A critical challenge in state estimation is maintaining consistent positioning across this boundary:

The Gap: Outdoors, Global Navigation Satellite Systems (GNSS) dominate; indoors, terrestrial systems (Ultra-Wideband/UWB, WiFi, Bluetooth Low Energy/BLE) take over.
The Failure Point: The building boundary (transition zone) is where both systems degrade simultaneously. GNSS visibility drops due to obstruction, while terrestrial systems suffer from limited line-of-sight (LOS) and multipath effects as the robot enters or exits.
The Missing Link: While individual domains (GNSS or indoor positioning) have mature error models, there is a lack of a unified, measurement-level view of how heterogeneous ranging technologies behave relative to each other across spatial zones. Current fusion approaches often lack agreed-upon handover protocols or robust error models for this specific transition regime.

2. Methodology

The paper presents a preliminary empirical assessment using the HYMN (HYbrid Multi-technology Navigation) dataset.

Experimental Setup:
- Location: An industrial facility in Torgau, Germany, featuring a 44m × 18m hall with a wide driveway gate, creating a natural indoor-outdoor transition.
- Ground Truth: A Leica TS16 total station provided millimeter-level precision measurements to establish a local Cartesian frame.
- Technologies: Four systems operated simultaneously:
  1. GNSS: NovAtel PwrPak7 (GPS L1 and Galileo E1).
  2. UWB: ZigPos RTLS (Two-Way Ranging).
  3. WiFi: HPE Aruba 630 (Fine Time Measurement/FTM).
  4. BLE: Metirionic DMK-215 (Phase-Based Ranging).
- Data Processing: Timestamps were synchronized, and GNSS ECEF observations were projected into the local frame. Residuals ( $r = \rho - d$ ) were calculated by comparing raw measurements against the known ground truth distance.
Ranging Model:
The authors model the raw range observation $\rho$ as:
$\rho = d + b + \eta + \epsilon$
Where:
- $d$ : True geometric distance.
- $b$ : Deterministic bias (clock offsets, calibration).
- $\eta$ : Environment-dependent stochastic error (NLOS, multipath), typically positive and heavy-tailed.
- $\epsilon$ : Zero-mean measurement noise.

3. Key Contributions

Per-Zone Characterization: An empirical analysis of four time-aligned ranging technologies across three distinct zones (Outdoor, Transition, Indoor), focusing on measurement availability and residual behavior against millimeter-level ground truth.
Identification of the "Double Degradation" Regime: The paper identifies that the indoor-outdoor transition is the specific regime where satellite and terrestrial systems degrade simultaneously, creating a critical vulnerability for state estimators.
Measurement-Level Framing: A structured view of heterogeneous data designed to guide robotic state estimator design, aiming to bridge the GNSS-navigation and indoor-positioning research communities.
Public Dataset: The release of the HYMN dataset for further research into cross-technology fusion.

4. Key Results and Observations

A. System Availability

GNSS: Satellite count drops sharply from ~25–30 outdoors to zero indoors. The transition zone shows a steep gradient, with visibility collapsing within a few meters of the gate.
UWB: Availability increases from ~5 observations outdoors (limited by LOS through the gate) to the full anchor complement (10) indoors.
WiFi & BLE: Remain broadly available across all zones, with variability driven more by local anchor geometry and obstruction than by the zone itself.

B. Ranging Residual Characteristics

The paper analyzes bias, median, standard deviation ( $\sigma$ ), and the 95th percentile of absolute residuals ( $P95|e|$ ):

UWB:
- Behavior: Near-zero median bias in LOS conditions.
- Error Profile: Tight distribution indoors ( $P95|e| \approx 0.99$ m) but degrades significantly outdoors and in the transition zone due to NLOS excursions.
- Outliers: Large standard deviation indoors is driven by sparse, heavy-tailed outliers (likely target confusion), not bulk noise.
WiFi FTM:
- Behavior: Exhibits zone-dependent positive biases (ranging from 1.17m in transition to 5.94m outdoors).
- Error Profile: Moderate spread, but the transition zone showed the tightest distribution, while outdoor performance was the worst.
BLE (Phase-Based):
- Behavior: Consistently large positive bias (5m–12m) and the widest spread of all systems ( $\sigma \approx 12$ m outdoors).
- Cause: Phase calibration errors, IQ imbalance, and multipath fold into the residual in the same direction.
GNSS (Single-Frequency BSSD):
- Behavior: Stable positive bias (9m–13m) and $\sigma \approx 10$ m across all zones.
- Observation: Even indoors, where few satellites are visible, the bias remains consistent, though the sample size is small.

C. The Transition Zone

The transition zone is the most critical area. Here, GNSS visibility is marginal, and terrestrial anchors are viewed at unfavorable geometries (through the gate). The paper concludes that a state estimator assuming equal measurement quality across zones will fail precisely where it is needed most.

5. Significance and Implications

Estimator Design: The results rule out the use of a single Gaussian covariance model for any technology.
- UWB requires handling sparse, heavy-tailed NLOS outliers.
- WiFi/BLE require handling large, zone-dependent positive biases.
- GNSS requires handling moderate, stable biases.
- Recommendation: Estimators should use non-parametric measurement models (Bayesian filters) or robust factor-graph back-ends (M-estimators, switchable constraints) to handle these heterogeneous error structures.
Community Bridging: The paper highlights a linguistic and methodological gap between the GNSS community (focused on integrity, atmospheric modeling, and ambiguity resolution) and the indoor positioning community (focused on heterogeneous modalities and self-calibration). It suggests that concepts like "protection levels" and "integrity monitoring" should be transferred to indoor robotics, while robust optimization techniques should be adopted for urban-canyon GNSS.
Infrastructure: The findings support the move toward self-calibrating infrastructure, as manual surveying of every anchor is not scalable for mobile robots moving between sites.

Conclusion

The paper demonstrates that seamless robot navigation across the indoor-outdoor boundary cannot rely on simple handovers or standard Gaussian noise assumptions. Instead, it requires zone-aware measurement noise adaptation and integrity monitoring that accounts for the specific, non-Gaussian error signatures of each technology. The HYMN dataset provides the necessary empirical foundation to develop these next-generation fusion algorithms.