🤖 AI

Information-Aided DVL Calibration

This paper proposes an information-aided calibration (IAC) method that enhances the accuracy of Doppler velocity log (DVL) measurements for autonomous underwater vehicles by improving conventional Kalman filter-based calibration in GNSS-enabled environments and enabling effective self-calibration when GNSS signals are unavailable.

Original authors: Zeev Yampolsky, Itzik Klein

Published 2026-07-01

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Zeev Yampolsky, Itzik Klein

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

The Big Picture: The Underwater GPS Problem

Imagine you are driving a car, but you are underwater. You have a very accurate speedometer (called a DVL) that tells you how fast you are moving relative to the ocean floor. However, this speedometer isn't perfect; it might be slightly tilted, or it might say you are going 10 mph when you are actually going 9.8 mph.

To fix these small errors, you usually need to surface and check your speed against a perfect reference, like a GPS on the surface. This is like checking your car's speedometer against a radar gun on the side of the road.

The Problem:

The "Tilt" and "Scale" Issues: Even after checking, the speedometer might still be slightly off because it's installed at a weird angle (misalignment) or its internal math is slightly wrong (scale factor).
The "No-Signal" Trap: Sometimes, the underwater vehicle (AUV) needs to dive deep where GPS signals can't reach. If the vehicle dives before the speedometer is perfectly calibrated, it has to guess its speed. Over time, these small guesses add up, and the vehicle gets lost (navigation drift).

The Solution: "Information-Aided Calibration" (IAC)

The authors of this paper propose a new way to calibrate the speedometer. They call it Information-Aided Calibration (IAC). Instead of just waiting for a perfect GPS signal, they use "common sense" rules about how the vehicle moves to help fix the errors.

Think of it like this: If you are driving a car on a straight highway, you know for a fact that you aren't sliding sideways or bouncing up and down violently. Even if your speedometer is a bit broken, you can use the fact that "I am only moving forward" to help fix the broken math.

The paper introduces two main strategies:

1. When GPS Is Available (The "Extra Check" Strategy)

The Goal: Make the calibration even better than the standard method, even when you have a GPS signal.
The Analogy: Imagine you are calibrating a scale. Usually, you put a 10lb weight on it. The authors say, "Let's also assume that the scale isn't wobbling up and down." By adding this extra rule (that the vertical movement is zero), the computer can figure out the errors much faster and more accurately.
The Result: They tested this on real underwater vehicles. By adding this "zero vertical movement" rule to their math, they improved calibration accuracy by 20% compared to standard methods.

2. When GPS is Not Available (The "Self-Reliant" Strategy)

The Goal: Calibrate the speedometer when the vehicle is deep underwater and has no GPS signal at all.
The Analogy: Imagine you are in a dark room with a broken compass. You can't see North, but you know you are walking in a straight line. You also know that you are mostly walking forward, not sideways or jumping up and down.
The authors created a new math model that assumes:

The vehicle is mostly moving forward (Surge).
The sideways (Sway) and up/down (Heave) movements are tiny or zero.
The total speed is the speed of that forward movement.

By using these assumptions, the vehicle can "self-calibrate" its speedometer using only its own data, without ever needing to surface.
The Result:

They tested three different versions of this "self-reliant" math.
The best version (Model M7) didn't even need to know anything about the vehicle's sideways movement beforehand. It just used the total speed.
This method improved speed estimation by 35% compared to using a completely uncalibrated speedometer.

The "Magic" Models (M3 to M7)

The paper proposes five specific math models (labeled M3 through M7) to handle these situations:

M3 & M4: Used when GPS is available. They add the "no vertical movement" rule to make the standard GPS calibration faster and more accurate.
M5, M6, & M7: Used when GPS is gone. They use the "mostly moving forward" rule to let the vehicle fix its own speedometer errors while deep underwater.

The Bottom Line

The researchers tested these ideas using real underwater vehicles from the University of Haifa. They found that:

With GPS: Their new method makes the calibration 20% more accurate.
Without GPS: Their new method allows the vehicle to fix its own speedometer, resulting in a 30–35% improvement in knowing where it is going.

Why it matters:
This means underwater robots can stay underwater longer, navigate more precisely, and complete their missions (like exploring the ocean floor or checking pipelines) without getting lost, even if they can't pop up to check their GPS. It turns a "broken" speedometer into a reliable one using smart math instead of expensive hardware.

Technical Summary: Information-Aided DVL Calibration

Problem Statement
Accurate velocity measurements from the Doppler Velocity Log (DVL) are critical for the navigation solutions of Autonomous Underwater Vehicles (AUVs), as they correct the drift inherent in Inertial Navigation Systems (INS). Traditionally, DVL calibration—estimating deterministic, time-invariant errors such as scale factors and misalignment angles—is performed on the water surface using Global Navigation Satellite System (GNSS) signals as a reference. This process relies on Kalman filter-based approaches and requires specific maneuvers to ensure state observability. However, two significant limitations exist:

GNSS-Denied Environments: In scenarios where GNSS signals are unavailable (e.g., underwater or in specific surface conditions), conventional calibration is impossible, forcing the use of uncalibrated DVL data which degrades navigation performance.
Accuracy Limits: Even in GNSS-enabled environments, the accuracy of traditional Kalman filter-based calibration may be suboptimal.

The paper addresses the challenge of improving calibration accuracy in GNSS-enabled scenarios and enabling DVL self-calibration in GNSS-denied environments without relying on external references.

Methodology
The authors propose an Information-Aided Calibration (IAC) framework that incorporates non-holonomic constraints and velocity approximations into the Kalman filter estimation process. The approach is divided into two distinct scenarios, resulting in five modified Kalman filter models (M3–M7).

GNSS-Enabled Scenarios (IAC-G):
The authors adapt non-holonomic constraints, typically used in ground vehicles, to AUVs. They propose that during straight-line calibration trajectories, the vertical (Z-axis) velocity component in the DVL frame averages to zero, even with periodic vertical motion.
- Models M3 and M4: These models integrate this "zero Z-velocity" assumption as an information aid alongside standard GNSS measurements. M3 applies this to a baseline model with a scalar scale factor, while M4 applies it to a model with an X-axis specific scale factor. This modification effectively adds a constraint to the measurement update without altering the core Kalman filter parameters.
GNSS-Denied Scenarios (IAC-GF):
For environments without GNSS, the authors propose a self-calibration framework based on three assumptions:
1. Simplified Error Model: Utilizing only misalignment errors (omitting the scale factor to reduce state dimensionality).
2. Near-Zero Sway and Heave: Approximating the body-frame Y (sway) and Z (heave) velocity components as small, near-zero values.
3. Surge Approximation: Deriving the unknown surge (X-axis) velocity by leveraging the invariance of the velocity norm across coordinate frames ( $||\tilde{v}_d|| \approx \tilde{v}_{IAC-GF,x}$ ).
- Models M5, M6, and M7: These models utilize the approximated velocity vector derived from the above assumptions.
  - M5 and M6 require prior knowledge or heuristics to estimate the sway and heave components.
  - M7 represents a novel, fully standalone approach. It utilizes only the measured DVL velocity norm and the X-component of the transformed velocity, requiring no prior knowledge of sway or heave components. It operates with a three-state vector (misalignment errors only).

Key Contributions

IAC Framework: The introduction of an Information-Aided Calibration framework that improves DVL calibration in both GNSS-available and GNSS-denied environments.
Zero-Velocity Aiding: In GNSS-enabled scenarios, the proposal to use zero-velocity information (specifically the Z-component) to enhance accuracy and reduce calibration time.
GNSS-Free Self-Calibration: The development of a relaxed error model combined with a novel velocity assumption (M7) that enables DVL self-calibration in GNSS-denied environments, offering an alternative to using uncalibrated measurements.
Five Novel Models: The formulation of five specific Kalman filter models (M3–M7) tailored to different aiding scenarios and state formulations.

Results
The proposed models were validated using real-world datasets collected by the University of Haifa AUV "Snapir" in the Mediterranean Sea. The evaluation involved 15 trajectories and a cross-validation procedure testing 15 different error-term combinations (scale factors, misalignments, biases, and noise).

GNSS-Enabled Performance:
- Model M3 achieved an average improvement of 20.12% in velocity root mean squared error (VRMSE) compared to the best baseline model (M2).
- Model M4 achieved an average improvement of 6.65% over M2.
- The results demonstrated that M3 and M4 converge faster and achieve lower error rates than baselines, even with shorter calibration durations.
GNSS-Denied Performance:
- Compared to the uncalibrated baseline, the proposed IAC-GF models significantly reduced navigation drift.
- Model M7 (the fully standalone model) achieved the highest improvement, reducing VRMSE by approximately 35% on average.
- Models M5 and M6 achieved average improvements of roughly 31% and 27%, respectively.
- The results confirmed that the heuristic approximation of sway and heave components (based on the first two seconds of data) provides a reasonable estimate for self-calibration.

Significance and Claims
The paper claims that the proposed IAC frameworks address critical gaps in AUV navigation by:

Enhancing Accuracy: Improving calibration accuracy in GNSS-enabled environments by up to 20% without requiring complex hardware changes or additional sensors.
Enabling Operation in Denied Environments: Providing a viable method for DVL self-calibration when GNSS is unavailable, preventing the reliance on uncalibrated measurements that lead to rapid navigation drift.
Mission Reliability: By reducing navigation drift and improving velocity estimation, the approach supports higher-quality data collection and safer mission execution.
Paradigm Shift: The work introduces a novel information-aided paradigm for underwater navigation, demonstrating that constraints and approximations (such as zero vertical velocity or velocity norm invariance) can be effectively leveraged to replace external references in specific operational contexts.

The authors note that the approach relies on assumptions regarding DVL installation (e.g., Z-axis alignment) and motion dynamics (straight-line trajectories), which may limit applicability in highly dynamic or non-standard configurations. However, the validation on real-world sea data confirms the robustness and practical applicability of the proposed models.