The Technological Fix: Building a Resilient Alternatives to GPS (Part 2 of 3)

The State of the Technology

Deep neural networks, combined with our depth of multispectral imagery collected both intentionally and incidentally over decades, offer the promise of a far more robust system that combines terrain association and inertial navigation to provide accurate information on precise locations.

(AI Generated Graphic)

Inertial navigation, a key component of path integration in unmanned equipment, has long been a useful mechanism for maintaining positional accuracy—even for equipment with a strong Global Positioning System (GPS) signal. The accelerometers, gyroscopes, and supplementary systems provide a constantly updated estimated position, but their accuracy degrades over time as the degrees of drift error compound. The most common example is civilian use of GPS in a tunnel—the mapping system will continue updating the vehicle’s location for a while after entering the tunnel, but the longer the GPS receiver goes without a signal, the less certainty the mapping system has regarding the positional accuracy. Even the best machine learning corrected inertial systems drift by approximately 0.5% of the total distance travelled. Eventually, the positional certainty falls below the threshold that the mapping system requires to provide an accurate location.

The industry has already begun efforts to improve inertial navigation, including significant advancements in quantum positioning systems (QPS).[1] These advancements provide improved accuracy by sensing magnetic field variations, exponentially extending the efficacy of inertial navigation.[2] They are not foolproof, however, and are prone to inaccuracies, from unmapped anomalies to geomagnetic storms, among other concerns.[3] The systems currently in testing, as described here, have the secondary benefit of being entirely passive, meaning they generate no electromagnetic signature, therefore eliminating a concern for military planners.

Mesh Network of Multi-Sensor Positional Systems

U.S. Air National Guard Emergency Management Airmen, Airmen First Class Valan Khalil, assigned to the ANG 3E9 Master Emergency Services Training (MEST) Course Class #014 participates in 5-level land navigation training led by the 188th Emergency Management Cadre at Fort Chaffee Joint Maneuver Training Center, Arkansas, May 21, 2026. The day-long exercise challenged 13 students from across the Air National Guard both physically and mentally while strengthening critical combat support skills, including compass navigation, map reading, terrain association, plotting courses, and the use of military GPS navigation systems. Conducted with support from Arkansas Army National Guard Range Control staff, the training enhances student confidence and readiness to perform emergency management tasks in operational and combat environments.

(U.S. Air National Guard photo by Tech Sgt. Patricia Teare)

Among the alternative navigation technologies available today, Simultaneous Localization and Mapping (SLAM) stands out as one of the most promising solutions for operations in degraded, denied, intermittent, or limited (DDIL) GPS environments. Long used in robotics, SLAM allows a platform to map its surroundings while simultaneously determining its own position within that map—an approach that provides accurate, independent navigation without reliance on satellite signals. The Army has already demonstrated success with TerraSLAM, a monocular-vision system developed in partnership with Carnegie Mellon University.[4] TerraSLAM fuses real-time optical sensing with preloaded geospatial information to produce accurate positional data using only a single camera, minimizing sensor requirements and simplifying integration.

TerraSLAM’s efficiency extends beyond its sensing approach. It operates effectively on lightweight processors, enabling affordable, distributed fielding across formations. The system’s accuracy—sufficient for both ground and aerial assets—has been validated through controlled testing, and its open-architecture design enables compatibility with select unmanned aircraft system, permitting autonomous missions even when GPS is unavailable. Just as units load communications fills or mission data, they can preload TerraSLAM with existing Army geospatial datasets, turning those holdings into an operational advantage. The navigation data generated can also be retained for after-action review, supporting refinement in evolving environments. While the system has yet to undergo full testing at PNT Assessment Experiment (PNTAX), DDIL Integrated Environment Supporting Experimentation and Learning (DIESEL), or All-domain Persistent Experiment (APEX), its demonstrated performance makes it a strong near-term candidate for field experimentation and rapid adoption.

Beyond SLAM, next-generation DDIL navigation will depend on integrating inertial measurement units (IMUs), artificial intelligence (AI)-driven path integration models, and multispectral terrain association. Path integration, inspired by discoveries in neuroscience, allows systems to estimate location based on motion and orientation data without external signals. When fused with AI-enabled terrain recognition, these models create a robust framework for continuous navigation that is inherently resilient to jamming and spoofing.

One approach to navigating in a DDIL GPS environment is to field a distributed mesh of positional sensors that fuses data from multiple, diverse sources. In practice, this would involve equipping personnel, vehicles, and unmanned systems with suites of inertial measurement tools—gyroscopes, accelerometers, barometers, and thermometers—that can continuously capture motion, velocity, altitude, and environmental shifts. On their own, such sensors are vulnerable to cumulative error, but when networked and cross-referenced, a mesh can compare readings across nodes, dampening error growth and creating greater resilience than any single unit could provide.

An industry partner and test flight lead, operates the Quantum Base program for a Vector AI unmanned aircraft system during Ivy Sting 5 on Fort Carson, Colorado, March 4, 2026. The capability supported Ivy Sting 5 by feeding persistent aerial surveillance into next generation command and control systems, enhancing situational awareness and decision-making abilities across the battlefield.

(U.S. Army photo by Pfc. Jacob Cruz)

The power of this system is multiplied when coupled with terrain association via multispectral imaging. Cameras capable of observing in visible, infrared, and other bands can match local terrain features against a library of reference imagery, similar to how terrain contour matching was once used to guide cruise missiles. Unlike the binary map models employed by Tomahawk missiles in the Gulf War, AI promises the potential to compile 2D images to create far more effective 3D models for quicker, more accurate association.

In a contested space environment where GPS signals may be unavailable or unreliable, the real-time fusion of inertial data with terrain-derived positional fixes enables forces to maintain accurate location awareness. Importantly, a mesh-based architecture allows for the distribution of error correction: one node that achieves a high-confidence terrain match can propagate that adjustment across the network, raising the overall accuracy and confidence level for all nodes in the system. This cooperative localization is particularly relevant for ground forces that maneuver in formations and need synchronized positional data without depending on vulnerable external signals.

When coupled with drones, manned aircraft, and known point-reference locations, the model can achieve a degree of accuracy as great as GPS. Combined with multispectral communication and targeting models, this mesh is capable of maintaining the targeting accuracy necessary to sustain conflict at a level below total war.

AI-Enabled Multispectral Map Models

While real-time sensor fusion provides immediate positional confidence, the effectiveness of terrain association hinges on the quality of the reference imagery. AI, particularly deep learning techniques, can greatly expand the value of decades of multispectral imagery already collected across the globe. By training models on historic image sets, AI systems can learn to normalize variations in lighting, season, vegetation, or atmospheric conditions and produce unified terrain models that highlight the stable, distinctive features most useful for navigation.

Maj. Gen. Patrick Ellis, 4th ID Commander (seated), prepares to depart for one of the areas of operations on May 6, 2026, during Ivy Mass, the culminating risk reduction event to validate Next Generation Command and Control (NGC2) systems across live and constructive environments at the Division level. Ellis’ Stryker served as the main “hub” as part of NGC2’s hub and spoke capabilities, with hubs hosting compute resources to provide local applications in Denied, Disrupted/Degraded, Intermittent, and Limited (DDIL) environments. The commanders’ hub connects to the various spokes; spokes are variants of capabilities integrated into multiple combat vehicles that can access applications on hubs for real-time on the move command and control.

Such models would not remain static. AI can continuously ingest new data from satellites, aerial platforms, and even field-deployed sensors, updating terrain libraries in near real time. The result is a living, adaptive map model that evolves as the physical environment changes—whether from construction, deforestation, seasonal shifts, or combat destruction. For terrain association tasks, this means higher correlation confidence and reduced false matches, even in regions where surface features change over time.

Moreover, AI can stratify features by persistence and reliability, weighting those that remain consistent (mountain ridges, major riverbeds, and urban cores) higher than transient ones (crop lines, snow cover). When combined with inertial path integration and a sensor mesh, these adaptive models allow positional data to be continuously recalibrated against a robust, current, and confidence-ranked map. For warfighters, this represents not only an alternative to GPS but also a system that could, under certain conditions, offer higher resilience by relying on features that cannot be easily jammed or destroyed.