The battlefield equation that dominated 2022 and 2023 — lots of cheap FPV strike drones versus layered air defenses — is mutating. On one axis you have mass produced, low cost first person view drones and loitering munitions. On the other axis you have dedicated electronic warfare to blind, jam, or misdirect those effects. In 2024 we are seeing a third axis develop: low‑cost onboard autonomy driven by computer vision and visual‑inertial navigation that reduces reliance on radio links and GNSS. That change matters because the defensive playbook that relied on simple jamming or signal suppression no longer covers the whole problem set.

How each side stacks up right now. Russian forces continue to deploy a suite of jammers and tactical EW nodes that can shape a cone of degraded satellite navigation and RF control over contested approaches. The operational effect is real: GPS interference has been credited with reducing the effectiveness of some GPS dependent Western munitions and forcing adjustments in employment. At the same time Ukrainian forces have pushed FPV drones, guided loitering munitions, and increasingly autonomy modules to keep strikes effective even when links are degraded. The result is not wholesale superiority for either side but a shifting set of tradeoffs that commanders must manage.

What the autonomy layer actually brings. Small, inexpensive onboard mission computers plus trained computer vision models are being integrated into attritable strike platforms to give them two practical capabilities: 1) the ability to navigate when GNSS is denied by matching onboard imagery against stored references or by fusing camera data with IMU (visual‑inertial navigation), and 2) the ability to perform terminal recognition or cueing so the drone can steer to hard or moving targets without continuous operator inputs. Those techniques have mature academic and engineering foundations and were demonstrated in research and industry releases through 2024. In short, visual navigation and image matching are not sci fi. They are usable methods to maintain mission effect when RF channels go dark.

Field reports and industry moves. By mid 2024 Auterion and several smaller vendors publicly described mission computers and packages intended to provide these capabilities at low cost and with field testing in Ukraine. Those solutions aim to let commodity and custom platforms continue to reach targets when GPS is jammed or operator links are weak. At the same time, reporting from the front shows Ukraine continuing to rely heavily on FPV strike craft for tactical effects and to experiment with autonomy where it improves reliability. This is not a single silver bullet. It is a pragmatic mix: keep doing what works in permissive corridors, add autonomy where jamming and range force a different approach, and prioritize simpler resilience for mass produced attritable systems.

Where EW still wins and the remaining failure modes. Autonomy helps, but it does not neuter EW. Visual navigation needs distinguishable terrain features, acceptable weather and lighting, and enough processing power for the inference required. Visual odometry and image matching perform poorly over homogenous terrain, over featureless snow or water, and at altitudes or speeds where ground features blur. If defenders can degrade optical sensors with obscurants, smoke, dust, or directed infrared countermeasures, or if they physically alter expected reference signatures, the autonomy margin shrinks. And autonomy that relies on preloaded reference maps can be defeated by deliberate camouflage, decoys, or rapid environmental change. Finally, autonomy introduces new signatures and behaviors for defenders to exploit; for example predictable flight profiles during visual homing.

Tactical recommendations from the frontline perspective. For Ukrainian operators and their allied advisors the practical checklist is: focus autonomy where it buys the most return on investment. Use visual inertial navigation and terminal recognition on long range loitering munitions and high value missions where link loss would otherwise abort the strike. Keep FPV manual control for close in, highly maneuverable attack runs where a human pilot still outperforms current models. Invest in quick model retraining workflows so recognition models learn local vehicle paint schemes, camouflage patterns, and signature variances. Pair autonomous strike kits with simple counter‑EW reconnaissance: use reconnaissance drones to locate and suppress jammers before committing high value guided rounds. Finally, accept that attritable quantity still matters; autonomy increases per unit effectiveness but does not replace mass when the objective is area denial.

Defensive adaptations to watch for. If I were advising Russian or any defender I would expect three countermoves: 1) more integrated multispectral obscurants and camouflage to deny visual cues, 2) greater effort to locate and physically destroy autonomy enabling infrastructure such as staging sites and reprogramming points, and 3) improved use of decoys and signature management to increase false positives for onboard classifiers. Jamming remains useful to shape the battlespace and to increase attrition of link dependent systems, but defenders will broaden their approach as attackers adopt onboard autonomy.

Operational limits and ethical observations. Greater autonomy reduces some tactical friction but raises target discrimination, collateral risk, and legal questions. Autonomy that simply stabilizes flight or performs navigation is one thing. Autonomy that selects lethal targets without human review is another. From a tactical engineering perspective, the near term wins are in navigation robustness and improved terminal control, not full autonomy of lethal decision making. Commanders and engineers need to keep those roles clear while deploying new capabilities at scale.

Bottom line. As of mid 2024 the contest between Ukrainian AI enabled drones and Russian EW is a classic iterative cycle. EW drives innovation toward GNSS and link resilient methods. Those methods work in many realistic conditions. In turn defenders will adapt with multispectral denial, signature manipulation, and target hardening. For practitioners the immediate imperative is pragmatic: field test simple autonomy modules, harden sensors and datasets against spoofing and obscurant tactics, and retain a mixed fleet where manual FPV, semiautonomous guidance, and mass production all have roles. Expect the cat‑and‑mouse game to continue, with incremental but operationally meaningful changes rather than a single deciding breakthrough.