Semi-Autonomous Drone Development in Ukraine 2025–2026: AI-Guided FPV and Strike Systems

The EW Imperative

In 2022 and early 2023, standard FPV drones operating on 5.8GHz video links were devastatingly effective on the Ukrainian battlefield. By 2024, Russian electronic warfare systems had achieved kill rates of 70–80% against these standard drones in heavily contested sectors — jamming the video link blinds the operator; jamming the GPS removes positional awareness; jamming the control link disables command input.

Ukraine's options were:

1. Frequency hopping: Spread-spectrum and frequency-hopping radio links raise the jamming difficulty. Deployed broadly. Partially effective but Russia adapts.
2. Fiber optic: Connecting drone to operator via physical fiber spool — unjammable but limits range (typically 10–15km) and adds mechanical complexity.
3. AI autonomy: Onboard processing that allows the drone to navigate and track targets without continuous communication link — jamming the link doesn't matter if the drone doesn't need it.

All three are being deployed. The AI autonomy approach is the most technically ambitious and strategically significant evolution.

What Is Semi-Autonomous?

A semi-autonomous combat drone operates in a spectrum between fully human-controlled and fully AI-controlled:

• Human-controlled baseline: Operator sees live video, controls all movement via RF link. Highly effective in low-jamming environment. Vulnerable to EW.
• Semi-autonomous (AI-assisted): AI handles navigation, obstacle avoidance, and target-lock tracking. Operator retains attack decision authority. Drone continues mission even if link is intermittent or blocked by terrain.
• Supervised autonomous: Operator approves a target pre-launch; the drone navigates autonomously to a grid area, identifies the target, and waits for human confirmation (or times out). Further reduces required link bandwidth.
• Fully autonomous (LAWS): AI identifies, selects, and engages targets without human decision. Ukraine publicly states it does not deploy fully autonomous lethal systems under that definition.

AI Capabilities Onboard

Modern semi-autonomous military FPV drones in the Ukrainian context feature combinations of:

1. Visual odometry: The onboard camera tracks scene features frame-to-frame to estimate drone motion without GPS — essentially computing position from optical flow. Effective in good visibility; degrades in smoke, darkness, or featureless terrain.
2. Terrain-based navigation: Pre-loaded aerial imagery of the target area. Onboard vision model correlates live camera view with the reference map to determine current position (visual SLAM — Simultaneous Localization and Mapping).
3. Target classification and tracking: Deep learning models trained on tank, APC, truck, helicopter signatures. Once an operator designates a target type pre-launch, the onboard model identifies and locks onto matching targets in the final kilometer. Models run on small AI accelerator chips (various including Ambarella, Qualcomm snapdragon edge chips).
4. Lock-after-designate terminal mode: Operator flies manually until target is in frame, "locks" the target with a button press. AI then maintains track and adjusts flight path as the target moves — similar to a heat-seeking missile's seeker logic but for optical lock on vehicles.

AI Drone Technology Comparison

Ukraine Development Programs

Ukraine has an extraordinary advantage in semi-autonomous drone development: a concentrated IT sector with deep AI/ML expertise (Kyiv and Lviv had significant AI research communities pre-war), combined with immediate access to real-world testing environments and experienced military operators who can provide feedback. Programs include:

• Saker Scout / Aerorozvidka AI program: Ukraine's drone reconnaissance unit has incorporated AI-assisted target detection and classification for reconnaissance drone imagery, identifying armor and vehicles automatically from video feeds
• Brave1 defense technology hub: Ukraine's Ministry of Digital Transformation and Ministry of Defense joint initiative to accelerate AI drone development, connecting civilian AI companies (many Ukraine-based) with military requirements
• UA Dynamics and other startups: Multiple Ukrainian defense startups developing AI-guided munitions, AI loitering drones, and computer vision targeting systems, several with Western venture and government backing
• Palantir partnership: US data analytics company Palantir has been deeply involved in Ukraine's battlefield AI, including target identification systems that could be integrated with autonomous strike platforms

Fiber Optic as Interim Solution

While AI autonomy is the long-term answer to jamming, fiber optic tethered FPV drones provide an interim solution for the most heavily contested environments. A fiber optic cable connects the drone to the ground control station via a rapid-unspooling lightweight spool mounted to the drone. Advantages:

• Impossible to jam electronically
• No RF emissions from control link (harder to detect/localize operator)
• Low latency video (better than compressed digital RF video)

Limitations: ~10–15km operational range before fiber weight and resistance limit practical performance; single-use spool (cost per mission increases); mechanical risk if spool tangles or severs. Despite limitations, fiber-optic FPV drones became an important capability in 2024–2025 for attacking heavily EW-defended positions.

AI Target Classification

One of the most operationally relevant AI capabilities for drones is automated target recognition (ATR) — identifying enemy vehicles in real-time video from an overhead or approach angle. Deep learning convolutional neural networks (CNNs) can be trained on thousands of labeled images of tanks, APCs, trucks, artillery pieces, and then run continuously on embedded edge-AI hardware (e.g., NVIDIA Jetson Nano/Orin at ~10W power draw) to classify and track targets in the video stream.

In practice this enables:

• Operator sees a target; presses "lock" — AI tracks it through maneuvers, smoke, and partial obscuration
• Reconnaissance drone automatically flags vehicle detections in footage, reducing analyst review time from hours to minutes
• Loitering munition can be given a waypoint search area and instructed to strike the first vehicle matching a category encountered

Key challenge: Distinguishing Russian and Ukrainian vehicles that are often similar types. Misidentification risk (fratricide) is the primary argument for maintaining human decision authority. Ukraine has reportedly invested significantly in solving this problem with visual recognition based on markings, antenna patterns, and other discriminators.

Russia's AI Drone Development

Russia has lagged Ukraine significantly in AI drone capabilities, for several reasons:

• Semiconductor access: Western export controls have severely restricted Russia's access to high-performance AI accelerator chips. Many Russian AI systems depend on Chinese suppliers as alternative, but highest-end chips are also restricted from China to Russia
• Software development ecosystem: Russia's AI research community has been significant, but the war's brain drain (emigration of tech workers), sanctions-driven software tool restrictions, and military secrecy culture have impeded rapid commercial-to-military AI adaptation
• Shahed optical guidance: Russia has incorporated optical correlator terminal guidance into some Shahed-136 variants — an onboard camera compares live imagery to a stored reference image of the target, making final approach independent of GPS and resistant to jamming. This is a meaningful capability, though less sophisticated than adaptive deep-learning targeting
• FPV AI chips: Reports and analysis of captured Russian FPV drones have identified Chinese AI visual processing chips incorporated into some variants, providing basic target-lock tracking in the terminal phase

Ethics and Legal Framework

International humanitarian law (IHL) requires that attacks discriminate between combatants and civilians (principle of distinction) and that expected civilian harm not be excessive relative to military advantage (proportionality). The application of these principles to AI-guided lethal systems is actively debated at the international level.

Ukraine has publicly maintained that its AI-assisted drone systems retain human decision authority for the terminal attack decision — a "human on the loop" standard. Whether this is implemented consistently in all operational conditions is difficult to independently verify, but it represents the stated policy. The International Committee of the Red Cross (ICRC) and the UN have called for international regulation of autonomous weapons systems (LAWS), and over 80 nations including Ukraine have expressed support for binding international rules.

March 2026 Status

The trajectory in 2026 points toward AI-assisted navigation and targeting becoming the baseline for advanced Ukrainian FPV and strike drones, with human supervisory authority retained but the operator's required active engagement reduced to key decision points. This evolution is driven by operational necessity — EW-saturated environments simply make human-piloted RF FPV increasingly ineffective at achieving hits.