Australia's Q-CTRL has run quantum gravity navigation at sea with satellite positioning excluded from the processing chain. In March 2026, the company sailed a 29-meter research vessel through the Coral Sea north of Cairns, Australia, matching gravity measurements taken onboard against existing maps over an 83-kilometer route. Suppressing inertial navigation errors without any external signal marks a step forward in a field that has largely relied on simulation studies or GNSS-assisted gravity surveys until now. However, the "one nautical mile" and "more than 10x" figures touted in the company's announcement do not describe the experimental results across the entire track. Tracing the numbers in the preprint reveals both how far the technology has come and how far it still has to go.
How cold atoms stopped inertial drift
Ships estimate their current position using an inertial navigation system (INS), which combines accelerometers and gyroscopes. While an INS operates without external radio signals, tiny sensor biases accumulate over time. In this evaluation, an INS without gravity-based correction drifted significantly within about six hours.
Q-CTRL's device is an atom interferometer that drops cooled rubidium-87 atoms and reads vertical acceleration from matter-wave interference. The roughly 30-liter sensor head also housed a classical accelerometer. The atom interferometer's measurement axis used here was limited to a single vertical axis, with horizontal motion and the ship's overall behavior handled by classical sensors and the IMU. The classical side manages the ship's fast movements and wide measurement range, while the atomic side serves as a low-frequency reference that stabilizes accelerometer bias. The quantum sensor alone did not determine the ship's position.
The stabilized acceleration data was combined with navigation IMU data. The software matched gravity anomalies reconstructed onboard against ocean gravity maps derived from satellite observations, correcting the INS's position, velocity, and attitude. This approach uses slight regional variations in gravity—caused by mass distribution near the surface—as landmarks that emit no radio signal.
While satellite altimetry can produce wide-area gravity anomaly maps at sea, publicly available magnetic maps have many gaps in observational data. This difference in map availability is part of why gravity was chosen as the reference.
In the navigation evaluation, GNSS position, velocity, and attitude data were excluded from the processing chain entirely—covering inertial calculations, gravity reconstruction, map matching, and state correction. The recorded GNSS track was used only afterward, as ground truth to measure error once processing was complete. This separation is what makes the test novel.
The "one nautical mile" claim doesn't apply to the full 83km
At the end of the roughly six-hour, 45-nautical-mile (83km) route, the uncorrected INS's position error reached 26km. GravNav, corrected using the gravity map, showed a terminal error of 4.1km—an improvement of about 6.3 times. At least along this specific route, the experiment directly compares the effect of adding gravity assistance to identical initial conditions and IMU data.
Meanwhile, GravNav stayed within about one nautical mile (1.852km) only for the first 70km of the route. The one-nautical-mile band shown in the paper's figure is a distance benchmark, not the root-mean-square error across the full 83km. By the endpoint, the error had widened to 4.1km.
In its August 27 announcement, Q-CTRL stated that it maintained one-nautical-mile accuracy during the mission and exceeded navigation-grade backup systems by more than 10 times. GPS World and Interesting Engineering largely echoed this framing. Against the company's summary of "one nautical mile over 83km, more than 10x improvement," the preprint limits the one-nautical-mile accuracy to the first 70km and reports the improvement at the 83km endpoint as roughly 6.3x—from 26km down to 4.1km. Furthermore, the comparison baseline is not satellite positioning itself but an INS without GNSS assistance.
This numerical discrepancy does not reverse the technology's overall success. The experiment does confirm that gravity assistance curbed the otherwise unbounded growth of error and kept it around one nautical mile for most of the route. Still, interpreting the result as "completing the full 83km at one-nautical-mile accuracy" overstates the finding.
The 300-meter gravity survey was a separate experiment
The vessel-based testing period lasted one week.
The same quantum gravimeter was also tested in a survey mode that referenced GNSS. Along a 280km coastal route, the team used a mechanically gimbaled configuration twice and a strapdown configuration—where the sensor is fixed directly to the hull—twice, for four passes total. They additionally surveyed a 145km loop route three times, encountering conditions up to Sea State 4 at maximum. This is separate data processing from the GNSS-free 83km navigation evaluation, and survey results were not fed into the navigation system.
In the strapdown measurements along the coastal route, within-run stability at 1,200-second integration was 0.67 mGal, with a 0.16 mGal average difference between the two passes. Shortening integration to 300 seconds brought stability to 0.93 mGal. On the loop route, the sensor repeatedly captured gravity anomalies of 15 to 40 mGal at a spatial resolution of about 300 meters. Given that the satellite map used had a half-power wavelength of about 16km, the spatial scale detected by the sensor was roughly 1/50th of that.
What was actually measured here is the repeatability of localized surveying by the quantum gravimeter. Figures like 300 meters or 0.16 mGal cannot be reinterpreted as positional accuracy for GNSS-free navigation. The paper itself acknowledges that existing classical strapdown gravimeters have reported numerically superior figures, such as 0.64 mGal. Differences in spatial bandwidth, routes, and correction methods also mean this isn't a simple performance ranking.
In a 56-hour stationary measurement, long-term drift referenced to the atomic sensor was about 1/70th that of the classical accelerometer alone. However, this was a stability test conducted while docked—not a result from 56 hours of navigation. The authors list multi-week closed-loop navigation as a priority for future work.
The distance from first public demonstration to operational hardware
The findings were reported by authors affiliated with Q-CTRL, led by Patrick J. Everitt. The manuscript is a v1 preprint submitted to arXiv on August 26, 2026—not a peer-reviewed paper, and it has no journal listing or DOI. The authors describe this as the first public demonstration of GNSS-independent map-matching navigation spanning the entire processing chain, using a mobile quantum gravimeter. No independent replication has been reported at this time.
This comparison ran uncorrected INS and GravNav from identical initial conditions, making it a genuine experiment measuring correction effects on this specific dataset. Still, the scope is limited to one vessel, one sea region, and roughly six hours. This was not a randomized trial, and it cannot support a general causal claim that the same effect would occur with different maps, sea conditions, or vessel types. It also needs to be distinguished from past simulation results.
Position error depends on map resolution as well as sensor performance. The satellite gravity map used here has a half-power wavelength of about 16km for cumulative anomaly signal power, meaning some stretches lack sufficiently fine landmarks. If actual measurements could pick up features at the 300-meter scale, that opens a path toward building higher-resolution maps from shipborne surveys—but navigation accuracy won't improve until such maps are prepared in advance.
The device's size also remains undetermined. The roughly 30-liter figure refers only to the sensor head. Calculating full volume would require counting the lasers and control electronics in an adjacent rack as well. The combined total including vacuum systems and power supply has not been disclosed. Removing the need for a gimbal and forgoing dedicated temperature control both ease deployment requirements, but this does not demonstrate dimensions suitable for mounting on autonomous vessels or submersibles.
What this test confirmed is that a GNSS-free processing chain—integrating a quantum gravimeter and classical sensors into an INS and matching against existing maps—can suppress inertial errors in real sea conditions. If independent parties can replicate multi-week closed-loop navigation, demonstrate error bounds in regions with sparse mapping, and disclose full volume and power consumption including peripheral equipment, GravNav will move closer to becoming a practical backup solution for situations where satellite signals are unavailable.