GNSS‑Denied UAV Autonomy: Sensor Fusion, Swarm Resilience and Mission Continuity in Contested Environments.
Interview on Autonomous UAV Operations Without GNSS
Miguel Ángel de Frutos Carro – Director and CTO at UAV Navigation-Grupo Oesía
When GNSS is degraded or denied, what does your autonomy stack fall back on, and how gracefully does that degradation happen in practice?
Our approach to autonomy in GNSS-degraded or denied environments is fundamentally based on multi-sensor fusion as a baseline, rather than attempting to replace GNSS with a single alternative. At UAV Navigation-Grupo Oesía, we treat robust inertial navigation combined with intelligent sensor fusion as the core of the system architecture.
By tightly integrating inertial sensors with complementary sources—such as air data systems, magnetometers, and, when available, visual navigation or SoOP systems—our estimation algorithms are able to maintain reliable navigation performance not only during temporary GNSS outages, but also under prolonged or even permanent GNSS denial. In purely inertial mode, the system typically exhibits a drift on the order of ~4% of the distance flown, which already enables meaningful mission continuation.
When aided by our Visual Navigation System, this drift is reduced to approximately ~1% in unknown environments, and can be effectively eliminated in areas that have been previously mapped or revisited. This highlights an important principle of our architecture: performance degrades progressively and predictably, rather than collapsing abruptly when GNSS is lost.
A key aspect of this graceful degradation lies in the system’s ability to perform continuous real-time integrity assessment. Our navigation algorithms are designed to detect and classify behaviors consistent with jamming, spoofing, or other NAVWAR techniques, allowing the estimator to reject unreliable inputs and prioritize trusted sources dynamically. This ensures stable and bounded performance even in highly contested electromagnetic environments.
Importantly, this is not a theoretical capability. It is a field-proven solution (TRL 9), currently deployed in real-world missions. Our systems have been tested against dedicated military counter-UAS and electronic warfare systems operated by multiple armed forces, demonstrating the ability to maintain navigation performance and complete missions under active interference.
That said, we do not consider GNSS-denied navigation a solved or static problem. We continue to invest heavily in expanding the robustness of our autonomy stack through additional aiding sources and technologies. These include advanced visual navigation techniques, terrain-referenced navigation (TRN), preloaded maps, and alternative positioning methods such as signals of opportunity (SoOp) or radio-based navigation.
Overall, our philosophy is clear: rather than relying on any single sensor or technology, we build layered, redundant navigation architectures that degrade gracefully, maintain operational continuity, and evolve continuously to address emerging threats in increasingly complex environments.
Swarm coordination assumes some level of inter-vehicle communication. When the RF environment is contested or jammed, how do you maintain coherent behavior across a formation — or do you shift to a different operational mode entirely?
Swarm coordination does assume some level of inter-vehicle communication. In our case, for tight formation flight, a communication link is required to guarantee safe and coherent behavior. While this link does not need to be perfectly continuous, any sustained degradation or loss forces a transition to safer, more separated formations, as the precise relative position of other vehicles can no longer be reliably guaranteed.
It is also important to frame this discussion in the correct operational context. We are referring to high-performance fixed-wing UAVs, operating at relatively high speeds and in dynamic environments—not to small, low-altitude, quasi-stationary platforms. This has a direct impact on both the communication geometry and the effectiveness of potential interference.
From a radio-frequency perspective, air-to-air links between closely spaced aircraft benefit from favorable propagation conditions, including short distances, line-of-sight, and the possibility of using directional transmissions with sufficient power. Compared to typical air-to-ground links, which are generally more exposed and easier to disrupt, these characteristics provide an inherent increase in robustness. However, it is essential to remain realistic: these links are not immune to jamming. A sufficiently capable or well-positioned adversary can still degrade or deny communications, which is why operational concepts cannot rely on uninterrupted connectivity.
For this reason, our approach does not depend on a single communication mechanism. We are actively developing complementary capabilities to ensure safe and coherent behavior in degraded environments.
One key line of work is vision-based relative awareness. Using onboard sensing and image processing, each UAV can detect, track, and maintain awareness of nearby aircraft. This capability enables a safety layer independent of communications, allowing for collision avoidance and basic formation awareness even when links are degraded or temporarily lost.
In parallel, we are investing in Free Space Optical Communications (FSOC) as a high-bandwidth, low-probability-of-intercept and low-probability-of-detection (LPI/LPD) solution. FSOC offers a communication channel that is inherently difficult to disrupt from the ground, although maintaining precise alignment between fast-moving platforms remains a technical challenge. We are currently collaborating with leading technology providers in this domain and positioning ourselves to demonstrate this capability in flight in the near term.
Overall, our philosophy is pragmatic: tight formations rely on communication and will safely degrade when it is lost, while multiple complementary technologies—directional RF links, vision-based tracking, and optical communications—are being integrated to extend operational robustness. This ensures mission continuity under realistic conditions, without relying on assumptions of perfect or persistent connectivity.
UAV Navigation-Grupo Oesía Real Swarm Flight, Vision Tests
Attritable platforms impose hard constraints on onboard compute. Where do you draw the line between what runs on the vehicle and what has to be offloaded or pre-computed, and how does that affect autonomy behavior in a denied environment?
Attritable platforms impose very real constraints not only on onboard compute, but also on power, cost, and producibility, which forces clear architectural trade-offs from the outset. These are systems designed to be expendable or to operate over a limited lifetime—typically well below 10 flight hours—and include a broad range of platforms such as target drones, decoys, loitering munitions, and other attritable UAV concepts.
Within this context, the key design principle is to strike the right balance between cost and capability, ensuring that the system is sufficiently effective to accomplish its mission while remaining scalable and economically viable.
From an autonomy perspective, we draw a clear boundary around what must always remain onboard and independent of any external connectivity. The vehicle must be capable, at all times, of maintaining safe and controlled flight, monitoring its own health, and executing the core elements of the mission. This includes guidance, navigation and control, safety-critical logic, contingency management, and the autonomy functions required to complete the mission even in the absence of communications. In denied environments, where links may be degraded, intermittent, or entirely unavailable, these capabilities cannot rely on reach-back or external support.
As part of this onboard capability, the system incorporates GNSS-independent navigation and multi-sensor estimation, ensuring that even under jamming, spoofing, or complete signal loss, the platform can continue operating in a predictable and bounded manner. The objective is not only to complete the mission when possible, but also to guarantee that, if the situation falls outside expected conditions, the system can transition safely to a predefined degraded or contingency mode.
Within these constraints, onboard processing is reserved for functions that have a direct and immediate impact on mission execution. This includes, for example, real-time image processing for terminal guidance or local perception, where latency, determinism, and robustness are essential. These are capabilities that must reside onboard because any dependency on external computation would introduce unacceptable risk in contested environments.
By contrast, higher-level and computationally intensive tasks such as mission planning, route optimization, map preparation, or AI model training are typically handled offboard or pre-computed prior to deployment. The vehicle is therefore equipped with what it needs to execute the mission reliably, rather than to perform heavy computation or learning during flight.
UAV Navigation-Grupo Oesía Manufacturing Line
At the same time, our architecture allows for a flexible distribution of more advanced decision-making capabilities. Through interfaces such as the Mission Control Computer or onboard mission processors, adaptive behaviors and AI-driven decision-making can be implemented onboard, shared across multiple platforms in a decentralized manner, or supported externally when communications allow it. However, this flexibility never compromises the fundamental principle that each platform must remain safe and operationally coherent on its own, even in complete isolation.
Finally, it is important to recognize that in attritable systems, engineering decisions are inseparable from industrial considerations. These platforms are designed to be produced at scale, often under significant time pressure, which means that both hardware and software architectures must be optimized not only for performance, but also for manufacturability, repeatability, and supply chain robustness. Internal processes such as calibration, testing, and quality control must be adapted accordingly to sustain production rates without compromising reliability.
Overall, our approach is based on a layered and pragmatic architecture: critical autonomy is always onboard and independent, advanced capabilities are modular and flexibly distributed, and all design decisions are grounded in the need to deliver systems that are not only operationally effective, but also scalable and sustainable in real-world conditions.
Eastern flank operational context has presumably informed your development priorities. Without getting into anything sensitive, what has real-world contested-environment feedback changed about how you think about the software stack?
Real‑world feedback from contested environments, particularly along the eastern flank, has reinforced a key idea for us: modern conflicts evolve faster than any single technology cycle. Advanced navigation systems progress rapidly, but so do counter‑UAS systems and electronic warfare capabilities designed to defeat them. As a result, technological excellence on its own is not sufficient. What truly matters is the ability to continuously adapt the software stack based on recent, real operational data coming directly from the field.
This has driven us to think of the software stack not as a static product, but as a living system supported by agile development and deployment processes. Close and direct collaboration with customers and operators allows us to receive timely frontline feedback, analyze emerging threats, and rapidly develop, validate, and deploy improvements. The objective is to shorten the loop between detection of a new challenge, engineering response, and return to operational service, often within extremely compressed timeframes.
In this context, it is more important than ever to remain true to the philosophy that forms the cornerstone of UAV Navigation-Grupo Oesía: listening to users’ actual needs in order to develop our systems. Operators working in contested environments provide insights that cannot be replicated in laboratories or simulations alone. By structuring our teams and workflows to absorb this information efficiently, we can deliver solutions that are not only technologically advanced, but also operationally relevant, reliable, and aligned with the realities faced in modern conflicts.
UAV Navigation-Grupo Oesía Flight Testing Area
Where do you see the genuine engineering limits right now — the problems that aren't solved by faster processors or better sensors, but require a different approach entirely?
Today, the real engineering limits are no longer primarily about sensors or processing power. In many areas, we already have technologies that work in isolation. The challenge lies in making them operationally viable at scale, in contested environments, and within realistic industrial and time constraints.
In navigation, for example, operating without GNSS should no longer be seen as an advanced capability, but as a baseline design requirement. The difficulty is not proving that GNSS-denied navigation is possible in a controlled scenario, but sustaining precision, continuity, and robustness in a repeatable, scalable, and cost-effective way. Solutions based on highly sophisticated sensors or increasing compute power may demonstrate excellent performance in specific cases, but if they cannot be produced, deployed, and maintained at scale, they do not address the real problem.
This is also why, despite the promise of emerging technologies such as gravimetric or quantum magnetic sensing, their practical adoption remains limited today. Beyond the sensor itself, these approaches depend on the availability, maintenance, and accessibility of high-quality environmental datasets, such as anomaly maps, which require significant and continuous effort to generate and update. In many cases, the bottleneck is not the sensor, but the data infrastructure behind it.
However, the deeper limitation is not purely technological—it is systemic. Traditional defense development models, based on long, sequential acquisition cycles and fully defined requirements from the outset, are increasingly misaligned with the pace at which technology and threats evolve. In this context, pursuing a theoretical “100% solution” often leads to systems that arrive too late or are no longer adapted to operational reality.
What is needed is a shift towards more agile, iterative development approaches, with shorter cycles, closer interaction with end users, and a willingness to accept that an 80% solution deployed early and continuously improved can be far more valuable than a perfect solution that never materializes in time. This also requires shortening the full chain from concept to deployment and enabling faster integration of new technologies as they mature.
Another key limitation lies in interoperability and industrial collaboration. Increasingly, no single company can master the full technological stack required for these systems. The future lies in ecosystems of highly specialized players, each contributing their core expertise, connected through well-defined interfaces and standards. In this context, competitiveness is not about building monolithic solutions, but about enabling cooperation between competent partners to deliver integrated capabilities.
Finally, as autonomy and AI become more central, access to high-quality datasets is emerging as a critical constraint. Training, validating, and deploying advanced algorithms requires large volumes of representative data, which are often difficult to obtain in defense contexts. Unlocking this bottleneck—through better data-sharing frameworks or structured data collection strategies—will be as important as advances in algorithms or hardware.
In summary, the real limits today are not solved by adding more processing power or better sensors alone. They require a different approach: rethinking development models, embracing modular and interoperable architectures, investing in data as a strategic asset, and aligning technological innovation with operational and industrial realities.
