Lavender Harvesting Robot: How to Replace a Sickle with Sensors

In the previous article we broke down how a weeding robot can free Crimean lavender plantations from weeds. The story got a response — and almost everyone who wrote back asked the same question: “What about harvesting? Wielding a sickle is even harder than wielding a hoe. Are you taking that on too?”

We are. And here’s why lavender harvesting is an even more grateful task for robotization than weeding. If weeding is about “weeds spoil the harvest”, harvesting is about “you literally have two weeks to collect everything, otherwise the essential oil will evaporate into the air”.

A two-week window

Every essential oil agronomist in Crimea knows the main rule: lavender must be harvested within 10–15 days of mass flowering. Not a month, not a season — two weeks. Here’s why:

Essential oil concentrates in the glands on the flower’s calyx. As the flower ages, the oil oxidizes and its yield drops by 30–50%.
After each rain you must wait at least 24 hours — until the sun “dries” the plant and oil accumulation resumes. In Crimea’s 2026 climate this means that of the two-week window, you have only 8–10 real work days.
Processing (distillation) must begin within 2–3 hours of cutting, otherwise the oil evaporates. That’s why fields are deliberately located next to the plant — every extra kilometer of transport hurts the quality of the raw material.

In 2026, about 1,100 hectares are under essential oil crops in Crimea, of which 770 are fruiting plantations, mostly lavender. By various estimates, that’s over 800 hectares that need to be harvested in “yesterday” mode. Manual sickle, mountainous terraces of Bakhchisaray district, a crew of 30–40 people — and prayers that the weather holds.

Why existing mechanization is not the answer

There are already lavender harvesting combines in the world. The most famous:

CLIER (France) — trailed combine, cuts tops at a fixed height, immediately into a container. Productivity — 1 ha in 3 hours, but only on flat fields.
MCL 2R — side-mount on a tractor with conveyor to trailer. Operated by joystick from the cab. Works, again, on steppe areas with level terrain.
HT-Tom (Italy) — motorized spindle head, embraces the bush from two sides and cuts. Light (42 kg), but requires manual feeding of each bush.

Sounds not bad until you remember Crimean specifics. The best plantations are in the foothills: Bakhchisaray, Simferopol districts. Slopes up to 12–15°, narrow terraces, rows of imperfect geometry. Neither CLIER nor MCL can drive in there — too heavy, require a tractor. And here the sickle enters again.

But the main limitation isn’t even the terrain. All existing combines cut “as is” — without distinguishing which bushes have already finished blooming, which are just entering peak, which are still green. And for Crimean varieties (Sineva, Vdala, etc.) selectivity is critical: you need to cut inflorescences at the 50–70% bloom phase, leave the rest to ripen. The combine can’t do that. The robot can.

What’s already being done in the world: lavender as a proving ground for selective harvesting

Lavender is a grateful crop for research teams: cute convex blue bushes, contrasting color, easy-to-measure yield. So lavender is used to refine selective harvesting technologies, which then become systems for strawberries, blueberries, and grapes.

Three notable projects of the past two years:

Michigan Technological University (USA). Professor Jung Yun Bae’s team won a $423,300 grant from the Michigan Department of Agriculture to develop a multi-robot platform for lavender harvesting. The concept: a central “manager” allocates bushes between robots, each robot drives up to its bush, finds inflorescences at peak phase, carefully cuts and stacks in a basket. When the basket is full — it goes to the unloading station and recharges itself. The 2025 prototype is “a team of robots working in the field without people”, as Bae describes.
IFIP / HAL (Europe). In 2025, the paper “Innovative Approach to Autonomous Lavender Harvesting Robot” was published, proposing a holistic architecture: row navigation + inflorescence recognition + custom cutting end-effector + multi-machine coordination through a central planner.
Dotnazanin (student project) — a small university team in the Netherlands assembling a lavender harvester prototype as a research stand for vision-based cutting. Also a characteristic trend: even student teams are taking lavender as a model task.

In Russia, at the level of institutes and startups, this niche is currently practically empty. Which, on one hand, is bad. On the other, it means the Crimean team that first brings a prototype to the field will get both patents and orders and beautiful case studies.

What a lavender harvesting robot needs to do: technical architecture

From all these projects and parallel developments in related crops (blueberries, strawberries, grapes), a clear stack emerges:

Active sensor placement on a lavender bush lives for 20–25 years, rows are aligned for decades. This is ideal for visual-inertial navigation: GPS gives rough position (±30–50 cm), and stereo cameras coupled with IMU lead the machine exactly along the row. Our basic “AgroBot” platform already has this layer — it navigated the vineyard with ±2 cm accuracy.

2. Computer vision for bloom phase

The main neural network task is determining the phase of each inflorescence. This is not “weed / not weed” like in the weeder. Here you need a classification of at least three classes: “green bud”, “peak bloom — cut”, “fading — skip”. Given that a single bush can have all three phases at the same time, the model must work in real time (10–15 frames per second) and in changeable Crimean lighting — from bright sun to golden hour before sunset.

3. Selective cutting end-effector

And here’s the most interesting part. Essential oil glands are fragile. Any crimping, any pressure — and the oil oxidizes. You need a cutting module that:

cuts, doesn’t crimp (oscillating knife or fast-rotating disk of minimal diameter);
takes the bunch not by the stem, but by the base of the inflorescence (a thin comb-stopper, like a beekeeper’s knife, but robotic);
doesn’t touch neighboring unripe buds;
lays the cut into a soft container without falling from a height.

From world experience, the most promising hybrid is a 6-DOF manipulator + a custom sickle-comb at the end. For Crimean terrain, it’s better to put two manipulators on the platform — to process two adjacent bushes at once per stop.

4. Logistics on the field

Cutting is half the battle. The cut needs to be quickly delivered for processing. The architecture here is: the robot carries a hopper for 30–50 kg of raw material on board, when full it automatically unloads into a mobile “collection point” (could be another robot or a regular trailer), and returns to the row. Ideally — a fleet of 3–5 robots works in the field, while the sixth shuttles between the field and the plant. This is already a classic multi-agent coordination task — what we learned in warehouse projects.

Economics: what the farm gets

Let’s count on a standard 50 ha of lavender.

Scenario	Productivity	Season ownership cost	Raw material quality
Manual sickle harvesting, 40 people	0.5–0.7 ha/day	2.5–3.5M ₽	High, but depends on weather and human factor
CLIER combine on tractor	3–4 ha/day	800K ₽ rental + fuel	Average: cuts everything indiscriminately, oil losses up to 25%
Selective harvester robot (fleet of 3)	4–6 ha/day	1.2–1.5M ₽ (with amortization)	High: cuts only peak phase, no crimping, no delay to processing

Fleet payback — 1.5–2 seasons. After that — pure savings and, more importantly, independence from seasonal workers, of which there are fewer and fewer in Crimean villages.

And there’s a non-obvious bonus. Selective harvesting gives +15–20% essential oil yield by cutting only inflorescences in the ideal phase. At a price of Crimean lavender oil of 8–12 thousand rubles per kilogram and an average yield of 60–80 kg of oil per hectare — that’s 200 to 400 thousand rubles of additional revenue per hectare. Just from raw material quality, the robot pays for itself even before the end of the first season.

What we plan to do at Tavrida Robotics

This article is not an abstract fantasy. We have a specific plan.

2026 (coming season):

Field research at 2–3 farms in Bakhchisaray and Simferopol districts. We collect a dataset of bloom phases on Sineva and Vdala varieties.
Cutting end-effector design. Goal — a prototype that cuts one bush without oil loss.
Computer vision model: first version of a bloom phase classifier.

2027:

Integration of the end-effector into the “AgroBot” platform (the same one we showed in the weeding article).
Field trials on 5–10 ha. Goal — to close the harvesting of one farm “from start to finish” with 2–3 robots.

2028:

Serial production of the harvesting module. Sale / lease to Crimean farms.

Why this matters beyond Crimea

Crimea is the birthplace of Russian lavender, but is far from the only region where it’s grown. Krasnodar Krai, Moldova, the North Caucasus — everywhere has the same problem: manual sickle, narrow harvesting windows, labor shortages. If the Crimean lavender harvester robot appears first — it will become an export product, and not only to CIS countries, but also to Bulgaria, France, Provence. Selective harvesting technologies are now under active development, and our trump card is practical field experience in a real farm, not a laboratory prototype.

What we need from you

If you grow lavender, sage, rose or other essential oil crops — we need your fields. Literally. One visit to your farm in July, when harvesting is underway, will give us more than a month of lab work. We’ll bring our equipment, shoot a dataset, discuss the pain points and timelines. In exchange — we’ll tell you how the robot will work, and lock in your priority for the first-line serial machines.

While we’re talking — harvesting is going with a sickle. Until we talk — it will continue to go with it for a long time.

Lavender should reach distillation within 2 hours of cutting. Not through a sickle, not through sweat, not through fatigue, not through “need to make it”.

Tavrida Robotics is a Crimean engineering company. We develop service and agricultural robots. The first article in this series — “Lavender Weeding Robotization”. Contact us via the form on the website or Telegram @TavridaRobotics.