QUICK ANSWER
Without GNSS, an inertial navigation system drifts: the position error grows with the square of time when the bias sits on the accelerometers and with the cube of time when it sits on the gyros. With an Epson M-G370PDG0, a well-aligned system travelling at 15 m/s is off by about 0.4 m after 10 s, 5.5 m after 60 s and roughly 280 m after 5 minutes (1σ, first-order model). Set your own outage time, speed and sensor below.
01
How do I read the result?
The headline value is the horizontal position error as DRMS (distance root mean square): about two out of three runs end inside this radius and roughly 95 % inside twice the radius. The thin lines are single Monte Carlo runs with random bias and noise, the white line is the closed-form 1σ result. Both come from the same model and therefore have to agree, which makes the plot its own plausibility check.
The error budget shows which data sheet line limits you at the selected time. For short outages it is usually the attitude and velocity error at the moment GNSS is lost. After a minute or more the gyro bias takes over, because it enters with t³.
02
Which bias value should I enter?
The three buttons under “Residual gyro bias” are three different lines of the same data sheet. Bias instability (0.8 °/h for the M-G370PDG0) is the floor a navigation filter can reach while GNSS is available. Repeatability (36 °/h) applies if you calibrated once and switched the system on again. Initial error (360 °/h, 1σ over −40…+85 °C) applies without any calibration.
In the default scenario this changes the error after 60 s from 5.5 m to 87 m and to about 870 m. How well your fusion filter estimates the bias therefore matters more than a small difference between two sensors. More on these lines in How to read an IMU datasheet.
03
How do the four Epson IMUs compare?
Default scenario: 15 m/s, residual bias equal to the bias instability, roll/pitch error 0.01°, heading error 0.1° and velocity error 0.02 m/s at GNSS loss (all 1σ).
| IMU | after 10 s | after 60 s | after 5 min | 10 m reached after |
|---|---|---|---|---|
| M-G570PR20 | 0.40 m | 5.2 m | 194 m | 84 s |
| M-G370PDG0 | 0.41 m | 5.5 m | 277 m | 80 s |
| M-G366PDG0 | 0.41 m | 6.1 m | 393 m | 75 s |
| M-G330PDG0 | 0.41 m | 9.2 m | 919 m | 62 s |
Own calculation with the model described below and typical values from the Epson data sheets. For the first seconds all four are equal, because the state at GNSS loss dominates.
04
What is the model behind it?
Each horizontal axis is treated as one channel of a strapdown inertial navigation system. The table lists the terms in their short-time form.
| Error source | Growth for short times | Symbol |
|---|---|---|
| Accelerometer bias | ½ · ba · t² | ba in m/s² |
| Gyro bias on roll/pitch | ⅙ · g · bg · t³ | bg in rad/s |
| Angle random walk | g · N · √(t⁵ / 20) | N in rad/√s |
| Velocity random walk | K · √(t³ / 3) | K in (m/s)/√s |
| Roll/pitch error at GNSS loss | ½ · g · δθ · t² | δθ in rad |
| Velocity error at GNSS loss | δv · t | δv in m/s |
| Heading error at GNSS loss | v · t · δψ | v speed, δψ in rad |
| Yaw gyro bias | ½ · v · bg · t² | |
| Yaw gyro noise | v · N · √(t³ / 3) |
For longer outages the tool uses the complete single-channel solution with Schuler feedback (period 84.4 min), for example R · bg · (t − sin(ωst)/ωs) for the gyro bias. Independent terms are combined as root sum of squares, along-track and cross-track errors are combined to the DRMS. The Monte Carlo runs integrate the same error equations step by step with random draws.
- P. D. Groves, Principles of GNSS, Inertial, and Multisensor Integrated Navigation Systems, 2nd ed., Artech House, 2013 (INS error propagation).
- D. H. Titterton, J. L. Weston, Strapdown Inertial Navigation Technology, 2nd ed., IET, 2004 (system performance analysis).
- O. J. Woodman, An introduction to inertial navigation, University of Cambridge, UCAM-CL-TR-696, 2007.
05
What does the simulator not cover?
- The vertical channel, which is unstable without a barometer or another height reference.
- Temperature and vibration: bias changes with temperature, and acceleration shifts the gyro bias (18 (°/h)/G on the Epson models).
- Scale-factor and misalignment errors, which add up during manoeuvres.
- Earth rotation and Coriolis coupling between the channels.
- Aiding by odometer, barometer, magnetometer or camera, and the correlations between errors in a real Kalman filter.
Use the tool for sizing and for comparing sensors, not as a performance guarantee. The way from these numbers to a project error budget is described in GNSS-denied navigation.
QHow long can an IMU navigate without GNSS?
That depends on the error you can accept. In the default scenario an M-G370PDG0 reaches 10 m (1σ) after about 80 s and 100 m after about 3.5 minutes. Enter your own limit in the tool to get the time for your case.
QWhy does the error grow so fast after a minute?
A gyro bias tilts the computed horizon. Gravity then leaks into the horizontal axes as a false acceleration, and this acceleration is integrated twice. Position error from gyro bias therefore grows with the cube of time.
QDoes a better IMU help during a 10-second outage?
Hardly. In the first seconds the attitude and velocity errors at the moment GNSS is lost dominate. The sensor grade decides how well the filter can estimate these states beforehand and how fast the error grows afterwards.
QAre these values guaranteed?
No. They are typical data sheet values in a simplified model. Measure under your own conditions before you commit to a design.
EVALUATION KIT
Check the numbers on your own platform
For qualified projects we provide evaluation kits for the Epson IMUs as a free loan. Our engineers help you turn the simulation into an error budget for your system.