When GNSS Is Lost: How Does the IMU Keep Navigation Alive?

A vehicle is travelling on an open road with a stable GNSS position. As it enters a tunnel, the number of tracked satellites quickly decreases until the GNSS solution is no longer available.

However, the navigation output does not immediately disappear.

The system continues to report position, velocity, and heading. The trajectory remains smooth, and for a short period, it may still appear to follow the road correctly. Only as the vehicle travels farther through the tunnel does the estimated position begin to drift away from its actual path.

Where does this position come from after GNSS is lost?

A common explanation is that the IMU “takes over” when GNSS fails. This description is convenient, but it is not entirely accurate.

In a GNSS+INS system, the IMU is normally working before, during, and after the GNSS outage. It does not suddenly switch on when the satellite signal disappears. Instead, the navigation system temporarily changes from a process of prediction and correction to one that depends mainly on inertial prediction.

Understanding this difference is essential when evaluating what a GNSS+INS system can really do during signal blockage.

1. The IMU Does Not Know Where It Is

A GNSS receiver and an inertial measurement unit provide very different types of information.

GNSS estimates an absolute position by processing signals from navigation satellites. Under suitable conditions, it can tell the system where it is on the Earth.

An IMU does not provide an absolute position. It normally contains accelerometers and gyroscopes:

• Accelerometers measure specific force along three axes.
• Gyroscopes measure angular rate around three axes.

Figure 1. Gyroscopes describe rotational motion, while accelerometers measure specific force and linear motion along three axes.

Using these measurements, an inertial navigation system estimates how the platform is rotating and accelerating. It then propagates the previous navigation state forward to calculate changes in attitude, velocity, and position.

GNSS answers: “Where am I?” The IMU answers: “How am I moving from where I was?”

This distinction explains both the strength and limitation of inertial navigation. The IMU can produce high-frequency motion information without relying on external signals. It can continue operating inside tunnels, under bridges, between buildings, or in other places where GNSS signals are blocked.

However, every inertial calculation starts from a previously estimated state. Small errors in the measurements are carried forward and accumulate over time.

A useful analogy is to imagine walking with your eyes closed. GNSS is like opening your eyes regularly and checking your location against visible landmarks. The IMU is like estimating your movement by counting steps and sensing each turn. You can continue moving when the landmarks disappear, but the longer you walk without checking, the more uncertain your location becomes.

2. GNSS Is Continuously Correcting the Inertial Solution

During normal operation, a GNSS+INS system does not usually calculate two completely independent positions and simply choose between them.

The measurements are combined through a navigation filter. The INS propagates position, velocity, and attitude at a high update rate, while GNSS observations are used to constrain and correct the estimate.

Figure 2. The IMU predicts motion continuously, while GNSS observations correct accumulated drift through the navigation filter.

The IMU measurements contain small imperfections, including:

• Gyroscope bias
• Accelerometer bias
• Scale-factor errors
• Sensor noise
• Axis misalignment
• Temperature-related variation

Even a very small gyroscope bias can gradually create an attitude error. Once the estimated attitude becomes inaccurate, measured acceleration may be projected into the wrong direction.

An accelerometer bias can create a small velocity error. Because velocity is integrated to calculate position, that velocity error continues to develop into a growing position error.

When GNSS is available, the navigation filter compares the inertial prediction with GNSS-derived information. The difference allows the system to correct not only the current position and velocity, but also estimates of IMU bias and other internal error states.

The IMU provides continuous, high-rate motion prediction, while GNSS provides the external reference needed to control long-term drift.

3. What Actually Happens During a GNSS Outage?

When the vehicle enters a tunnel and GNSS tracking is interrupted, the inertial navigation calculation does not stop.

The system still has the most recently estimated:

• Position
• Velocity
• Roll, pitch, and heading
• IMU bias
• Sensor calibration states
• Platform motion state

It uses the incoming accelerometer and gyroscope measurements to continue propagating the navigation solution. This period is often described as inertial coasting, dead reckoning, or free-inertial navigation.

Figure 3. Before the tunnel, GNSS corrects the inertial solution. During the outage, the IMU keeps predicting and drift grows. After GNSS returns, the solution is progressively corrected.

Immediately After GNSS Is Lost

At the beginning of an outage, the solution may remain relatively close to the actual trajectory. This is because the system starts from a recently corrected state. The IMU bias estimates may also be well calibrated if the system completed proper initialization and had good GNSS observations before entering the blocked area.

The output can therefore remain smooth and continuous. However, smooth output should not be confused with guaranteed accuracy.

As the Outage Continues

Without GNSS updates, the system has no new absolute-position reference. Small residual sensor errors begin to accumulate:

• 1. A gyroscope error causes the estimated heading or attitude to drift.
• 2. The attitude error affects the direction in which acceleration is applied.
• 3. Accelerometer errors create incorrect velocity estimates.
• 4. Velocity errors accumulate into position errors.
• 5. The uncertainty of the navigation solution continues to increase.

The important question is not whether the system can still output a position, but how long that position remains accurate enough for the application.

There is no universal answer such as “an IMU can bridge GNSS loss for 30 seconds.” Outage performance depends on IMU sensor grade, initialization quality, GNSS accuracy before the outage, vehicle dynamics, temperature, vibration, installation, outage duration, aiding information, and the application’s accuracy requirement.

4. Why Different Platforms Behave Differently

Two systems using similar IMUs may perform very differently during the same length of GNSS outage. The reason is that performance depends not only on the sensor, but also on platform motion and the constraints available to the algorithm.

Road Vehicles

A road vehicle normally moves forward, remains close to the ground, and has limited lateral and vertical motion. The navigation system may use vehicle-motion constraints, and wheel-speed data may provide another estimate of travelled distance. These constraints can help limit some types of drift, although wheel slip, sharp turns, uneven roads, and incorrect configuration can reduce their effectiveness.

Precision Agriculture and Ground Robots

Agricultural machines and mobile robots may operate at lower speeds, but vibration, wheel slip, frequent turns, and uneven terrain can make inertial estimation difficult. Useful aids include wheel speed, non-holonomic constraints, reliable heading initialization, vibration isolation, stationary detection, and slip detection.

UAVs

A UAV can accelerate, rotate, climb, descend, and move sideways. There are fewer simple motion constraints than on a wheeled vehicle. During an outage, even a small attitude error may cause gravity or vehicle acceleration to be projected into the wrong axes, producing rapidly increasing velocity and position errors.

Stationary or Frequently Stopping Equipment

When the system can reliably determine that it is stationary, it may apply a zero-velocity update. Any calculated movement can then be used to estimate and reduce inertial error. The stationary state must be detected reliably; applying the constraint while the equipment is moving would introduce new errors.

5. When GNSS Returns, the System Must Rebuild Confidence

Leaving the tunnel does not necessarily mean that the navigation solution instantly returns to full accuracy. The receiver may begin tracking satellites again, but the newly available observations still need to be assessed.

The system may need to consider:

• Number and geometry of available satellites
• Signal quality
• Multipath near the tunnel entrance
• Consistency between GNSS and INS
• Size of the accumulated inertial error
• GNSS positioning status
• Whether RTK ambiguity fixing has recovered
• Whether the observations contain outliers

If the INS estimate has drifted and the first recovered GNSS position is affected by multipath, immediately forcing the navigation solution to the new coordinate may cause a sudden and incorrect trajectory jump.

A well-designed fusion algorithm therefore uses recovered GNSS information according to its estimated reliability. As valid observations continue to arrive, the filter can correct position, velocity, attitude, gyroscope bias, accelerometer bias, and other estimated errors. Recovery may be progressive rather than an abrupt step.

📘 Recommended Reading

How Multi-Frequency GNSS Eliminates Ionospheric Delay

Ionospheric delay is one of the major error sources affecting GNSS positioning. Learn how multi-frequency GNSS helps reduce ionospheric errors and improve positioning accuracy.