Why INS/DVL Alone Cannot Solve Long-Duration AUV Drift

Inertial navigation system (INS) positioning errors accumulate continuously with time — a fundamental physical limitation that makes long-term high-precision positioning impossible without external correction. The standard architecture pairs a Strapdown Inertial Navigation System (SINS) with a Doppler Velocity Log (DVL): the SINS integrates accelerometer and gyroscope measurements to estimate attitude and velocity, while the DVL provides ground-referenced velocity from acoustic Doppler returns off the seabed. Both sources compound errors, and without GPS — which is unavailable underwater due to electromagnetic signal attenuation — there is no periodic absolute fix to reset the accumulated drift.

A dual-monitoring architecture developed by Beijing Zhonghang Tianyou Technology Co., Ltd. (2026) addresses both error sources simultaneously. The system continuously monitors IMU acceleration variation and DVL wave-velocity quality fluctuation in parallel, performs independent analyses of each signal’s reliability, and applies a dual-determination decision logic to select the appropriate position correction scheme. A “path position credibility score” is maintained over successive time windows, triggering path stability correction only when drift accumulation has exceeded a statistically significant threshold — delaying rather than eliminating the fundamental drift problem.

Harbin Engineering University’s INS/DVL/LBL tight-coupling patent (2023) demonstrates this advantage directly. The architecture integrates INS, DVL, Long Baseline (LBL) acoustic positioning, and intelligent pressure sensors (IPS) in a tight-coupling Kalman filter, performing real-time attitude, velocity, and position error feedback correction. For deep-sea AUVs that cannot access DVL bottom-tracking returns during descent — because the seabed is beyond DVL range — Northwest Polytechnical University (2018) describes a bidirectional smoothing strategy: a forward Kalman pass during the mission is supplemented by a backward re-processing pass once the AUV surfaces and acquires GPS, correcting the entire submerged trajectory retroactively. This “forward + reverse” scheme reduces mission-average positioning error without requiring any in-mission surfacing events.

Acoustic Positioning as an External Drift-Correction Anchor

Because INS/DVL drift is unbounded over time, the principal engineering strategy for long-duration missions is periodic injection of absolute position fixes from acoustic systems. Unlike INS, Long Baseline (LBL) acoustic positioning does not accumulate errors, meaning it can bound SINS drift indefinitely within transponder coverage — a property that makes it the dominant external correction method across the patent dataset.

Sparse LBL: Large-Area Coverage with Minimal Infrastructure

Conventional LBL arrays require four seabed transponders whose positions must be precisely calibrated before a mission. CSIC 707 Research Institute (2018, updated 2020) directly addresses this deployment cost with a sparse LBL tight-coupling method that uses only two transponders yet achieves positioning performance equivalent to conventional four-transponder arrays. As the AUV approaches the edge of one transponder pair’s coverage, it deploys a new pair and uses a local sub-filter to estimate their positions — enabling seamless, rolling coverage for large-area, long-duration missions without restricting the AUV’s operational area to the acoustic array configuration.

Bearing-Only Passive Acoustic Positioning

Single-beacon passive acoustic positioning eliminates the need for multiple seabed transponders entirely. Southeast University’s bearing-only positioning patent (2019) places a single acoustic source on the seabed and computes the AUV’s geodetic position from the phase differences between signals received at bow and stern hydrophone arrays. By using phase difference (angle-only) rather than time-of-flight (range), the method is inherently immune to sound-speed irregularities that cause distance measurement errors. The AUV does not need to surface for position updates, preserving covertness — a critical requirement for military and sensitive survey applications.

“Bearing-only passive acoustic positioning is inherently immune to sound-speed irregularities that cause distance measurement errors — and the AUV never needs to surface, preserving covertness throughout the mission.”

Sound-Speed Uncertainty: A Fundamental Bottleneck

All acoustic range-based positioning methods are affected by uncertainty in the underwater sound speed profile — which varies with temperature, salinity, and depth. China University of Petroleum (East China) (2025) addresses this directly with a semidefinite programming method that constructs a pseudo-linear positioning equation set, applies semidefinite relaxation to convert the constrained least-squares problem to a semidefinite programming problem solvable via interior-point methods, and then refines the estimate through a weighted least-squares error correction step — simultaneously estimating AUV position and the unknown sound speed. A related 2025 patent from the same assignee uses DVL velocity measurements combined with periodic acoustic signal reception to construct a virtual long baseline array from a single beacon, addressing scenarios where no seabed infrastructure exists.

Machine Learning and Data-Driven Error Compensation

When sensors fail or environmental conditions degrade measurement quality, purely physics-based navigation systems lose accuracy rapidly. DVL failure is a critical single-point vulnerability in standard AUV navigation: if the DVL loses bottom-lock or malfunctions, the INS receives no velocity correction and drift accelerates dramatically. A distinct family of approaches uses machine learning to model, predict, and compensate for navigation errors without requiring additional hardware.

Dynamics-Based Velocity Models for DVL Failure

Ocean University of China (2019) addresses DVL failure with a dynamics-based velocity model trained on historical data incorporating propeller RPM, rudder angles, heading angle, and AHRS-measured accelerations and angular rates. When DVL anomalies are detected, the model-predicted velocity transparently substitutes for the DVL output, preventing the “large navigation deviations or even system paralysis” that would otherwise occur. Crucially, the training set is continuously augmented during normal operation, so the model improves as a function of each vehicle’s specific hydrodynamic characteristics over time — a form of on-board continual learning.

SVM-Based Dead-Reckoning Error Correction

Harbin Engineering University’s SVM-based error correction patent (2011) is an earlier landmark in data-driven AUV navigation. A Support Vector Machine is trained on paired records of sensor readings and GPS-measured ground truth from surface trials. The trained model is embedded in the AUV’s flight computer and continuously predicts the dead-reckoning error from real-time sensor inputs during submerged operation, providing real-time compensation without surfacing. The patent explicitly acknowledges that rule-based correction methods “are only accurate under specific conditions” and that SVM generalization across changing environments is the key differentiator — a finding that anticipates the broader shift toward learned navigation models now evident in more recent filings.

FastSLAM for AUV Docking Homing

Shenyang University (2025) describes a SLAM-based approach for the homing and docking scenario — where both AUV and docking station position must be estimated simultaneously. The method extends the FastSLAM 2.0 algorithm with a fuzzy Q-learning process noise adaptation scheme, addressing the common problem that “process noise statistical characteristics are difficult to accurately measure in the underwater environment” — a factor that otherwise causes SLAM filter divergence and gross positioning errors over long missions. The fuzzy Q-learning component adaptively tunes the process noise covariance in real time, maintaining filter consistency even as environmental conditions change.

Seabed Coherence Mapping and Cooperative Multi-AUV Positioning

Two additional strategies extend drift correction to scenarios where neither acoustic infrastructure nor DVL is reliable: terrain-referenced navigation using seabed imagery, and cooperative positioning across AUV swarms. Both approaches are capable of providing absolute position fixes with no fixed seabed infrastructure — a capability that is uniquely valuable in denied environments or deep-ocean areas where transponder deployment is impractical.

Synthetic Aperture Sonar Coherence Map Navigation

Raytheon Company’s coherence map navigation system (PCT filing 2017; JP versions 2019, 2020) is the most technically distinctive approach in the dataset. The system acquires complex synthetic aperture sonar returns of the seabed during an ingress pass and generates a complex seabed image. During the return egress pass, locally generated complex images are compared against the stored ingress image using a normalized cross-correlation coefficient (NCCC). A maximum NCCC value indicates that the AUV’s along-track position has been found by map-matching, providing an absolute position fix with no acoustic infrastructure whatsoever. The system continuously monitors the NCCC trend to determine whether the AUV is converging to or diverging from the reference track — enabling closed-loop position correction along the entire egress path. According to IEEE standards on synthetic aperture sonar processing, coherence-based map matching is among the most robust approaches for terrain-referenced navigation in featureless deep-sea environments.

Cooperative Multi-AUV Positioning

Multi-AUV cooperative positioning leverages inter-vehicle acoustic ranging to provide external position corrections without any fixed infrastructure. Beijing Institute of Technology (2014) describes a fully distributed protocol in which each AUV periodically broadcasts its current position and a coordinate accuracy metric (lambda). When an AUV’s own accuracy falls below a threshold, or when at least three neighbors have higher accuracy, it updates its position using neighbor coordinates — effectively propagating high-accuracy positions through the swarm. The patent explicitly accounts for ocean current effects on positioning accuracy degradation over time.

CGG Services SA (US 2019; WO 2015; EP 2016) implements a commercial-scale cooperative positioning system for seismic survey AUV arrays. Each AUV detects acoustic signals emitted by neighboring AUVs, determines their relative positions, and performs corrective maneuvers to maintain a pre-planned formation. The formation geometry itself acts as a rigid positioning reference: because the planned inter-vehicle spacings are known, any AUV can compute its absolute position from neighbor ranges. Beijing Institute of Technology’s related 2014 patent uses inter-vehicle distance measurements via TDOA acoustic ranging to calibrate dead-reckoning tracking coordinates, “reducing tracking errors and extending the effective underwater working time.” Multi-sensor fusion between surface unmanned vehicles (USVs) and submerged AUVs is also addressed by Harbin Engineering University Sanya Nanhai Innovation and Development Base (2025), which coordinates USV-mounted acoustic positioning equipment with AUV navigation to provide continuous underwater correction without surfacing — an approach consistent with the heterogeneous vehicle architectures increasingly documented by WIPO in its annual technology trends reports on marine robotics.

Patent Landscape: Who Is Leading GPS-Denied AUV Navigation R&D

The patent dataset of more than 50 records filed between 2011 and 2026 across Chinese, US, European, Japanese, and PCT jurisdictions reveals a clear concentration of innovation activity in Chinese academic institutions, with a smaller but technically distinctive contribution from Western defence and commercial players. Understanding who holds IP in each sub-domain is essential for R&D teams assessing freedom-to-operate and partnership opportunities — a point underscored by the OECD‘s recent analysis of ocean technology IP concentration.

Harbin Engineering University is by far the most prolific assignee, with patents covering INS/DVL/LBL tight integration, SVM-based error correction, sliding-mode trajectory tracking, and multiple derivative institutions (Harbin Engineering University Sanya Nanhai Innovation and Development Base, Qingdao Innovation Center). Their work spans from 2011 to 2025, indicating sustained, multi-decade investment. Zhejiang University contributes advanced integrated positioning and control architectures, particularly the acoustic-buoy/EKF/MPC integrated framework. Southeast University focuses specifically on passive acoustic positioning without seabed infrastructure, with the bearing-only single-source and time-window positioning methods. CSIC 707 Research Institute addresses the operational scalability problem through the sparse LBL tight-coupling method.

On the Western side, CGG Services SA (France) holds a significant international patent family on cooperative multi-AUV formation positioning, with filings in the US, WO, EP, and MX jurisdictions — the only major Western commercial seismic company in this dataset with dedicated navigation IP. Raytheon Company holds the most technically distinctive approach: synthetic aperture sonar coherence map navigation, which requires no acoustic beacons of any kind. China University of Petroleum (East China) represents emerging innovation in sound-speed-robust single-beacon positioning via semidefinite programming. The breadth of assignees filing in multiple jurisdictions aligns with the global patent filing trends documented by EPO in its annual patent index for marine and autonomous systems technology.