Inertial Navigation and Sensor Fusion for Continuous Positioning

Positioning in marine and construction environments is rarely continuous when it depends on satellite signals alone. GNSS provides absolute position referenced to a global frame, but the signal is vulnerable to blockage, multipath, and interference. Inertial navigation systems provide continuous position and orientation by measuring motion directly, but their accuracy degrades with time when no external reference is available. Sensor fusion combines both, using the strengths of each system to compensate for the weaknesses of the other.

Why Continuous Positioning Is Difficult

The problem is not limited to open water. A survey vessel passing beneath a bridge, a crane boom blocking the sky view, a dredger working close to a quay wall, an ROV beneath a platform, and an excavator operating beside a stockpile all create conditions in which GNSS quality deteriorates without warning. In high-dynamic environments, the vessel or machine also changes attitude faster than a GNSS solution updates, so attitude must be interpolated between satellite epochs.

Two requirements therefore coexist. The system must deliver position at a rate consistent with the motion being measured, and it must maintain orientation accurately enough to resolve gravity correctly and to steer sensors, blades, and beams. Neither GNSS nor an inertial measurement unit satisfies both requirements independently.

Physical Basis of Inertial Navigation

An inertial measurement unit contains three orthogonal accelerometers and three orthogonal gyroscopes. The gyroscopes measure angular rate. Integrating angular rate yields changes in attitude. The accelerometers measure specific force, which is the combination of kinematic acceleration and the reaction to gravity. When attitude is known, the gravity component can be removed, leaving kinematic acceleration. Integrating acceleration once yields velocity, and integrating again yields position.

Inertial navigation is a dead-reckoning process. Position is obtained by integrating acceleration twice, and attitude is obtained by integrating angular rate once. Because integration accumulates error, inertial accuracy degrades with time unless an external reference is used to bound the drift.

Error Propagation and Drift

Accelerometer bias produces a position error that grows approximately with the square of time. A bias equivalent to one milli-g, applied without correction, generates roughly two metres of position error within one minute. Gyroscope bias produces attitude error that grows linearly, and because gravity is resolved using attitude, a small tilt error couples a component of gravity into the horizontal channels. That coupling produces an apparent horizontal acceleration, which itself integrates into position error.

In an unaided system near the Earth's surface, the horizontal error channels oscillate with a period of approximately 84.4 minutes, known as the Schuler period. This behaviour is not a defect; it is a consequence of the physics of a terrestrial inertial platform. It explains why inertial systems are inherently stable in the long term only when aided.

Grades of Inertial Sensors

• Navigation grade: Very low gyroscope and accelerometer bias, suitable for long coasting periods, typically large and expensive.
• Tactical grade: Moderate bias and drift, commonly used in hydrographic survey and hydrographic-grade motion sensing.
• Industrial and MEMS grade: Higher bias and drift, small and low cost, suitable for machine control and short-outage bridging when aided frequently.

Sensor grade determines how long the system can maintain specification after losing GNSS. It does not remove the need for aiding; it only changes how quickly the solution decays.

GNSS as an Aiding Source

GNSS provides position and velocity that are stable over long periods. Real-time kinematic positioning delivers centimetre-level horizontal accuracy, differential GNSS delivers sub-metre accuracy, and standalone positioning delivers several metres. Vertical accuracy is generally weaker than horizontal, which is significant for marine and earthworks applications that depend on elevation or depth.

GNSS also has rate and latency limitations. Solutions typically update at rates between one and twenty hertz, while inertial sensors update at one hundred hertz or more. Latency in the position solution must be modelled and compensated, otherwise dynamic errors appear between the position fix and the sensor it is intended to support.

GNSS and inertial navigation are complementary in the frequency domain. GNSS provides bounded low-frequency accuracy with noise and limited update rate. An inertial measurement unit provides smooth high-rate data with unbounded low-frequency drift. Fusion uses GNSS to estimate and correct inertial errors while using the inertial solution to smooth and interpolate between GNSS epochs.

Sensor Fusion Architectures

Loosely Coupled Integration

GNSS position and velocity are treated as measurements in the navigation filter. This architecture is simple, robust, and compatible with almost any GNSS receiver because it uses the receiver's final solution rather than raw observables. Its weakness is dependence on the receiver's own solution quality. When satellite geometry degrades or fewer than four satellites are tracked, the GNSS solution becomes unreliable and the filter loses its main reference.

Tightly Coupled Integration

Raw pseudorange and Doppler measurements are supplied directly to the filter. Because the inertial solution provides the geometry that the satellite measurements lack, the filter can continue to operate with fewer than four satellites. Tight coupling also improves rejection of multipath and degraded measurements, because individual observables can be weighted or discarded without losing the entire position fix. It requires access to raw GNSS measurements, which not all receivers provide.

Deeply Coupled Integration

Inertial data is fed back into the receiver's tracking loops to assist signal acquisition and tracking. This approach improves performance under weak signal conditions and high dynamics, and it shortens reacquisition time after obstruction. It is more complex and is generally reserved for specialised applications.

The Navigation Filter as the Integration Engine

A Kalman filter is the standard estimator for GNSS and inertial fusion. The state vector typically includes position, velocity, attitude, accelerometer bias, gyroscope bias, and, in tightly coupled designs, GNSS receiver clock terms. The filter predicts the state forward using inertial measurements and then corrects that prediction using GNSS or other aiding measurements. It also maintains an estimate of its own uncertainty, which determines how much each measurement influences the solution.

Filter configuration is a practical engineering decision. Process noise and measurement noise must reflect the real behaviour of the sensors. Over-optimistic settings produce an overconfident solution that ignores valid measurements; over-conservative settings produce a sluggish solution that rejects useful corrections. Innovation monitoring, which examines the difference between predicted and observed measurements, is essential for detecting faults and configuration errors.

Complementary Aiding Sensors

GNSS and inertial fusion is often extended with additional sensors that bound specific error channels.

• Doppler velocity log: Provides bottom-track velocity for underwater vehicles and constrains horizontal drift when GNSS is unavailable.
• Wheel odometry or vehicle bus speed: Constrains forward velocity for land vehicles and reduces drift during short GNSS outages.
• Magnetometer: Provides heading reference where the magnetic environment is stable, but is unreliable near steel structures, motors, and high-current cables.
• Pressure or depth sensor: Constrains vertical position for subsea vehicles and ROV operations.
• Acoustic positioning: Ultra-short baseline or long baseline systems provide absolute position for ROVs and towed bodies relative to a vessel or seabed array.
• Zero velocity update: When a vehicle is stationary, a zero-velocity measurement is applied to constrain drift. This is a simple and effective technique for land and amphibious equipment.
• Dual-antenna GNSS heading: Provides absolute heading independent of vehicle motion, removing a major error source for slow-moving or stationary vessels.

Marine Applications

Multibeam bathymetry depends on attitude as much as on acoustics. Beam steering, refraction correction, and outer-beam quality all require accurate roll, pitch, and heading at the moment of transmission. Vessel heave must be measured to correct the transducer depth, and GNSS vertical accuracy alone is generally insufficient for that purpose. A fused inertial solution provides the high-rate heave record needed for tide-independent vertical referencing.

Uncrewed surface vessels operate in shallow water and near structures where GNSS reception is intermittent. Continuous fused positioning allows the platform to maintain a track and keep acoustic data georeferenced across outages. The same logic applies to dredge guidance, where cutter or draghead position must remain reliable near quay walls and moored vessels.

Machine Control Applications

Earthmoving machines change attitude rapidly and generate vibration that affects both GNSS antennas and inertial sensors. Fusion of GNSS with an inertial measurement unit provides blade or bucket position at rates higher than the GNSS solution alone, which improves grading smoothness and reduces hydraulic lag error. During short outages, such as passing beneath an overhead structure, the inertial solution maintains an estimated position until GNSS returns.

On tracked machines, the inertial unit must be mounted rigidly and, where necessary, isolated from high-frequency vibration. Loose mounting or flexible bracketry introduces attitude error that no filter tuning can correct.

Optimization Strategies and Common Pitfalls

• Good practice: Allow a static alignment period before mobile operations so the system can initialise attitude and estimate gyroscope bias.
• Good practice: Verify lever arms between the GNSS antenna, the inertial measurement unit, and the reference point, and re-verify after any mechanical change.
• Good practice: Use dual-antenna GNSS heading where available, particularly on slow-moving vessels.
• Good practice: Monitor GNSS quality indicators and the filter's own uncertainty in real time, and treat rising uncertainty as an operational warning.
• Good practice: Log raw inertial and GNSS data so that post-processed trajectories can be generated for final deliverables.
• Good practice: Perform a known-point check before production and after any sensor change.
• Poor practice: Mounting the inertial unit on a vibrating, flexible, or thermally unstable structure.
• Poor practice: Ignoring lever arm errors, which create attitude-dependent position errors that increase with distance from the antenna.
• Poor practice: Relying on magnetometer heading near steelwork, machinery, or high-current cables.
• Poor practice: Resuming operations immediately after a long GNSS outage without re-alignment or verification.
• Poor practice: Assuming that inertial bridging can continue indefinitely. Position error grows with time and must be bounded by re-acquiring GNSS or another absolute reference.
• Poor practice: Accepting the real-time fused solution without post-processed verification on projects with tight vertical or horizontal tolerances.

Practical Implications for Geomak

Geomak applies GNSS and inertial fusion across hydrographic survey, dredging control, ROV and towed-body operations, and machine control for earthworks. The deliverable is not a single position value. It is a continuous, documented trajectory with an associated uncertainty statement, supported by calibration records, lever arm verification, and quality indicators from both the satellite and inertial sides of the solution.

For marine and construction projects in the Gulf, where heat, vibration, shallow water, and dense infrastructure all challenge satellite reception, a properly configured fusion solution is what allows positioning to remain continuous rather than intermittent. The engineering effort belongs in system design, calibration, and verification, not in post-hoc correction of a degraded trajectory.