Industries / Food, farming and forestry / Agriculture and farming / Nursery and horticulture
Picking and packing
Taking ripe fruit, vegetables and flowers off the plant, then grading, cooling, storing, packing and selling them, often across an ocean.
5
Milestones achieved, from 1912 to 2018.
3
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1912
Flowers sold by falling clock
Growers around Aalsmeer, in the Netherlands, set up two cooperative flower auctions: Bloemenlust held its first sale in the Welkom café on 4 December 1911, and the group that became the Centrale Aalsmeersche Veiling held its first in January 1912. Both came to sell by the "Dutch auction" clock, on which the price falls until a buyer stops it, a method Dutch vegetable growers had taken up in the late 1880s. The two merged in 1968 as Verenigde Bloemenveilingen Aalsmeer, which let growers from other European Community countries sell from 1972 and growers outside the European Union from 2006.
Credit
the Aalsmeer growers' cooperatives
1929
Apples kept in controlled air
At the Low Temperature Research Station in Cambridge, England, Franklin Kidd and Cyril West studied how temperature, carbon dioxide and oxygen change the speed at which fruit ripens. The first commercial apple store to use their "gas storage" method was built in England in 1929, according to Michigan State University horticulturist David Dilley. Controlled-atmosphere (CA) storage, which keeps fruit cold in air with less oxygen and more carbon dioxide, reached US apple stores in the early 1950s, and by the late 1980s US CA capacity for apples and pears exceeded 100 million bushels.
Credit
Franklin Kidd and Cyril West
1965
Flowers fly to market
On 18 October 1965 Edgar Wells shipped the first commercial load of cut flowers from Bogotá to Miami, a consignment worth $20,000. By the US Department of Agriculture's count, Colombia's cut-flower exports were worth $2.4 billion in 2025, Colombia supplies about 60% of flowers sold in the US, and the sector employs more than 200,000 people.
LessonAn industry that sends nearly 80% of its output to one country depends on that country's trade policy; a 10% US tariff from April 2025 was estimated by the industry to add more than $200 million a year in costs.
Credit
Edgar Wells and the Colombian growers who followed him
2002
A gas that pauses ripening
1-Methylcyclopropene (1-MCP) attaches to the fruit's receptors for ethylene, the gas that sets off ripening, and so slows it. Edward Sisler and Sylvia Blankenship of North Carolina State University found it; it was patented in 1996, and the US Environmental Protection Agency approved it in 2002. AgroFresh sells it as SmartFresh, which Growing Produce reports is used in 45 countries and on up to 70% of apples harvested in the US.
Credit
Edward Sisler and Sylvia Blankenship; AgroFresh
2018
Pepper robot needs new plants
SWEEPER, a European Union project led by Wageningen Research from 2015 to 2018, tested a sweet-pepper-picking robot in a commercial Dutch greenhouse on 262 fruits. It picked 18% of ripe peppers in the usual double-stem crop and 61% in a simulated single-stem crop, taking 24 seconds per fruit, against 6% and 106 seconds for the project before it, CROPS. The project wrote that a robot working beside human pickers would need under 10 seconds per fruit, about 50% of the fruit and under 1% losses.
LessonA picking robot did far better when the crop was grown to suit it, so the plant and the machine have to be designed together.
Credit
Wageningen Research and the SWEEPER partners (EU grant 644313)
Watch for
Snack-tomato robot picks 45%
inaho says its robot for snack (cherry-sized) tomatoes raised its harvest rate from about 15% to over 45% of the fruit and nearly doubled its speed to 20 kg an hour in a third season of trials at the Dutch grower Kwekerij Duijvestijn, so that four robots could cover about 1 ha. In November 2026 it reported that the robot worked in daily production at two unnamed growers in the Netherlands and Belgium from June to August 2026, and that one of them "has entered the adoption stage for 2027".
Achieved whena grower or an independent party publishes a full paid season of robot results: share of ripe fruit picked, kilograms an hour, and damaged fruit.
Claimed by
inaho, October 2025 and November 2026
Cherry-tomato robot in daily use
Tokuiten, a Japanese startup, says its harvesting robot entered full production use on 25 May 2026 at the company's own 2,000 m² organic cherry-tomato farm in Chita, Aichi Prefecture, after one unit harvested 31 kg a day in April 2026. Staff run it as part of their regular work, each session picks about half what one person picks, and all of the fruit is sold through normal channels. A six-robot fleet is planned for a 1 ha farm due in March 2027.
Achieved whena grower other than Tokuiten runs the robot for a season and the share of ripe fruit picked, kilograms per hour and damaged fruit are published.
Claimed by
Tokuiten, June 2026
Machine cuts and bunches chrysanthemums
TTA-ISO, Horizon Technologies and the grower Linflowers are building the HVC, a machine that cuts chrysanthemum stems in the greenhouse bed, sorts them into three or four grades with cameras and load cells (scales), and groups them into bunches of five. The partners aim to have it "fully validated in an operational chrysanthemum greenhouse" by March 2027, when the EU-supported project ends; no speed or cost figures are given.
Achieved whenthe machine runs in commercial production and a grower or outside party publishes stems per hour, grading agreement with human graders and damaged stems.
Claimed by
TTA-ISO, Horizon Technologies and Linflowers, March 2026
Open
01
Vase life from one photo
A phone app and a public set of photos that predict how many days a cut rose will last in a vase, from one photo taken at packing, checked against roses kept in water and scored every day.
The briefCredit · unclaimed02
Robot builds a mixed bouquet
A robot station that takes single stems from buckets and builds a hand-tied spiral bouquet to a written recipe (for example three roses, two alstroemeria and two stems of greenery), ready for binding.
The briefCredit · unclaimed
Open 01 · the brief
Vase life from one photo
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A phone app and a public set of photos that predict how many days a cut rose will last in a vase, from one photo taken at packing, checked against roses kept in water and scored every day.
Why it is still open
Vase life is still judged by keeping stems in water and checking them by eye each day for drooping necks, wilting and dry petals. The nearest prediction model, from Andong National University in Korea (Frontiers in Plant Science, 2024), used a hyperspectral camera (an IMEC SNAPSCAN that records 150 bands of light) on two cultivars, and the authors call for more cultivars and conditions. The same group's 2025 monitoring system watches roses in a chamber with 12 cameras and scales, one image a minute, and its vase-life estimate matched observation only moderately (r² of 0.66, on one cultivar). A search found no model or public dataset that predicts vase life from one ordinary phone photo.
Built with what exists
A recent phone camera and a simple light box with a color card; Label Studio for recording each stem's daily score; an open image model fine-tuned with PyTorch, or an open vision-language model run locally with ggml (see the Note "A first stable number for local AI"); cut roses bought from a wholesaler.
First step
Buy 50 stems of one rose cultivar, photograph each in the light box on arrival, and start a daily vase-life log with a fixed scoring sheet.
Done looks like
500 stems from at least three cultivars, each photographed on day 0 and scored daily until it fails, with the model's average error in days published beside a simple baseline (each cultivar's average vase life), and all photos and scores released under an open license.
Open 02 · the brief
Robot builds a mixed bouquet
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A robot station that takes single stems from buckets and builds a hand-tied spiral bouquet to a written recipe (for example three roses, two alstroemeria and two stems of greenery), ready for binding.
Why it is still open
Havatec calls its Twister "the first machine in the world that is able to make twisted flower bouquets automatically", but its turntables "are filled manually according to the entered bouquet recipe". FloraBot, a MassRobotics startup working with QuickFlora, said in 2019 that its robot arm could make "a typical dozen-roses arrangement in about four minutes" against 15–17 minutes for a person, with no test method given, and a 2020 report on its kiosk design gives no figures and no pilot. No later report of it in use was found. Labor is 50–60% of the cost of producing Colombian flowers, by the US Department of Agriculture's count.
agriculture-xprt.comquickflora.comfloraldaily.comagriexchange.apeda.gov.in
Built with what exists
Two SO-101 open-design arms working together, recorded and trained with LeRobot (see the Note "No download"); an open robot model such as NVIDIA GR00T 2 (see the Note "Five demonstrations per task") or Runway Praxis-1 (see the Note "Open robot model weights"); wrist and overhead cameras; a clamp that holds the bouquet as it grows.
First step
Build the bouquet clamp and record 50 remote-controlled single-stem placements with LeRobot.
Done looks like
20 filmed bouquets to one seven-stem recipe, with a published tally of stems placed correctly, broken heads or stems, dropped stems and seconds per bouquet, set beside a florist's time for the same recipe.