By: Dawn Zoldi
In autonomy, contested environments have moved from exception to baseline. VICTUS founder and CEO Jesse Hamel built his company, and its entire software architecture, to address this challenge. Instead of optimizing for blue skies, clear links and perfect GPS, VICTUS has engineered “synthetic autonomy” that can survive jamming, spoofing, clouds, water and urban canyons, what Hamel and his team call “contested autonomy.” This article traces how his operational career, technical choices and business model all converge around autonomy that still works even when everything else collapses.
From AC‑130 Gunships to Synthetic GPS
Hamel has deep roots in contested skies. He joined the U.S. Air Force, inspired to service by the events of 9/11, where he spent 20 years flying AC‑130 gunships, conducting research with the Air Force Research Laboratory, working on precision munitions and commanding an Air Force Special Operations drone squadron. Following his impressive military career, he completed work at MIT in machine learning (ML) and launched VICTUS from that ecosystem to connect his combat experience with state‑of‑the‑art autonomy research.
Throughout Hamel’s experiences, GPS fragility consistently arose as a “Level 5” pain point, like a “K-bar that’s stuck in the ribs.” He described watching mission after mission hinge on a navigation backbone that adversaries could distort or deny, from red‑flag exercises that emulated “everything is dark” to modern theaters like Ukraine, where electronic warfare (EW) has forced operators to use fiber optic cables, visual navigation and other workarounds. “The problem being,” he explained, “is that most of the autonomy software…forgot to solve the ‘Where am I?’” problem. As soon as you put a robot into real‑world conditions filled with solar flares, clouds, water or jamming, “the whole thing kind of crumbles,” he said.
That realization seeded VICTUS. Rather than accepting brittle autonomy that fails when GPS disappears, Hamel built a software that could synthesize its own reliable navigation backbone from onboard signals and learned artificial intelligence (AI) models, even when the constellation is lying or missing.
After validating his ideas in MIT hackathons and early projects, Hamel leveraged programs like Techstars Miami and other accelerators to pressure‑test both the business model and the technology with real customers, including U.S. Special Operations Command. Just two years strong, the company continues to gather momentum in both defense and commercial circles. Among other achievements, it has collected MOUs, accelerator slots and programs like NATO DIANA and Palantir’s DevCon Startup Fellowship that explicitly focus on autonomy in contested spectrum and GPS‑denied environments.
Contest As the Constant, Not the Edge Case
VICTUS exists to address environments where GPS‑centric autonomy fails: battlefield EW, dense cities, greenhouses, mines, tunnels, pipes, over‑water corridors and even commercial or high-end business aviation. Hamel characterized this “contested autonomy” as “where humans live.” We do not operate in sanitized labs, he noted. We operate in dirty, reflective, cluttered, adversarial environments, whether on a Ukrainian front line, a mine tunnel, a greenhouse complex or a congested capital city. If autonomy cannot maintain navigational truth there, it will never achieve the “five nines” reliability that unlocked cloud computing and other infrastructure revolutions.
According to Hamel, it starts with the fact that GPS signals arrive at the earth’s surface with the effective power of a light bulb, easily disturbed by reflections or buildings. Optical‑flow and vision‑dependent techniques collapse over water, sand, snow, glass or featureless terrain. He traces much of this fragility back to a single architectural assumption: the Kalman filter–centric guidance, navigation and control stack built in the 1960s and scaled globally after GPS was declassified in the late 1990s. That math works well when you have either clean GPS or reliable visual cues. It was never designed for the environments in which modern robots actually need to operate.
The result is a mismatch between how autonomy is deployed and where it is needed most. The same GPS unreliability that can cause an Uber or Lyft pickup to “spin around” outside a Vegas casino becomes an existential threat when multiplied across swarms, beyond visual line of sight (BVLOS) inspection fleets or drones operating under heavy EW. Hamel noted that RTK‑based drone shows, which rely on centimeter‑level GPS for hundreds of vehicles, can be brought down with a “$50 GPS jammer.” This frightening prospect illustrates how much of today’s autonomy still rests on a single point of failure. GPS should be used when available, he argued, but never treated as the single source of navigational truth.
Inside PhantomNAV: Machine‑Learned Navigation That Survives
VICTUS designed PhantomNAV, the company’s flagship contested‑environment navigation software, as the answer to GPS fragility. The company has organized its entire lifecycle, from modeling, simulation and ML training through sim‑to‑real validation and edge deployment, around the assumption that dust, haze, buildings, solar flares, urban canyons and electronic attack are normal. It built its solution from the hardest environments up.
“Our whole model,” Hamel said, “is to enable the producers of autonomy to have systems that work in these environments.” He sees VICTUS as part of a software supply chain that allows OEMs, integrators and governments to meet emerging contested‑autonomy requirements without pricing their platforms out of existence.
A pure‑software, hardware‑agnostic layer built to sit alongside existing navigation modalities (for drones, think: autopilots like PX4, ArduPilot and Pixhawk), PhantomNAV observes the system and steps in only when GPS becomes unreliable.
The process begins long before a robot flies. VICTUS models the specific edge system, whether a drone, ground robot or vessel, then places that model into a synthetic environment where it can experience the full spectrum of GPS‑denied conditions in accelerated time. A custom multi‑state machine learning (ML) approach, backed by a tailored data ontology, trains on inertial inputs such as accelerometer and gyroscope data, to learn how the system should move relative to gravity and other dynamics, even without external fixes.
VICTUS calls the output of that training “synthetic GPS”: a machine‑learned state estimator that produces a high‑confidence position stream from inertial data alone, compressed to run on lightweight edge compute like a Raspberry Pi 4. Once deployed on the platform, the software passively monitors the onboard Kalman filter and GPS quality. If GPS remains solid, PhantomNAV stays in the background. When GPS becomes jammed, spoofed or lost, and the error balloons from meter‑level accuracy toward kilometer‑scale drift, the ML state estimator feeds its synthetic GPS into the architecture in the format the autopilot expects to effectively “stand in” for the missing constellation.
Because it plugs into existing reference architectures and does not require new APIs or proprietary hardware, PhantomNAV can retrofit fleets without forcing OEMs into expensive sensor upgrades or exotic inertial systems. Hamel contrasts this software‑first approach with legacy alternatives such as “quantum” INS units quoted at roughly $80,000 per drone, an obviously untenable option for manufacturers targeting sub‑$2,000 airframes. With PhantomNAV, contested‑environment navigation becomes an AI‑driven software capability, not a hardware tax.
VICTUS’ architecture also becomes a force multiplier for human operators. By moving the “Where am I?” problem into a resilient onboard layer, PhantomNAV enables a single human to influence multiple drones from different manufacturers through tools like the Tactical Awareness Kit (TAK), without worrying that a brief GPS outage will send each platform wandering off course.
Where to Go To Deepen Your Contested Autonomy Playbook
VICTUS has rejected the traditional PNT sales pattern of incremental antenna upgrades and “magic boxes” with marginal real‑world impact. Instead, the company leans into what Hamel calls “integration as a service.” VICTUS takes customer platforms through a structured process: environment modeling, synthetic data generation, ML training, bench testing, sim‑to‑real gap closure and operational validation.
Once that integration is complete, the commercial model shifts to a straightforward software‑as‑a‑service (SaaS) structure, priced per device and designed to be “orders of magnitude cheaper than any other attempt” at GPS‑denied navigation. The compute footprint stays intentionally small, aligning with OEMs’ SWaP‑c pressures and enables PhantomNAV to ride along on modest hardware without eroding battery life or reliability.
For autonomy builders, policymakers and investors that want to dig deeper into how to survive in contested autonomy environments:
- Start with Dawn of Autonomy Episode 107 featuring Jesse Hamel for a section‑by‑section walkthrough of contested environments, synthetic GPS and the new “contested autonomy” category.
- Explore VICTUS’ site for more on its simulation‑driven training pipeline, reinforcement‑learning architecture and “orbit‑to‑seabed” ambitions for hardware‑agnostic navigation.
- Revisit Autonomy Global’s earlier feature, “Out of Stealth and Into the Fight: VICTUS Gives Autonomy a Backbone When GPS Fails” for additional detail on PhantomNAV, synthetic GPS and the company’s emergence from stealth, without re‑treading the ground you have just read here.
“If you’re making a robotic system that depends on GPS, it has no future in national security or serious industrial autonomy,” Hamel bluntly stated in summary. Systems that internalize contested autonomy and navigational resilience as first‑order requirements will define autonomy in the next decade.