Industries / Construction and property / Landscaping and lawn care
Tree care
Tree care, or arboriculture, is the pruning, treatment, risk checking and removal of trees in yards, streets, parks and campuses.
6
Milestones achieved, from 1901 to 2022.
1
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1901
Tree surgery in a book
John Davey of Kent, Ohio, published The Tree Doctor in 1901 and advertised his crews' services in it. He taught crews to scrape decayed wood out of a wound "much the same as a surgeon would adopt in scraping a diseased bone" and then fill the hole. One tree in the book took "over four tons" of Portland cement to fill, a treatment Davey said added one to two hundred years to the tree's life.
LessonFilling and painting wounds looked like good care, but later research showed it does not stop decay (see 1983).
Credit
John Davey, founder of The Davey Tree Expert Company
1930
Dutch elm disease arrives
Dutch elm disease is a fungus spread by bark beetles that blocks an elm's water-carrying wood. The USDA Forest Service dates its discovery in the United States to 1930, and a 2016 Forest Service workshop paper by Michael Marcotrigiano says it had "a devastating impact" on American elm in cities and along rivers. A breeding effort begun in the 1930s infected more than 20,000 elm seedlings; about 40 years later only 16 had survived, and none was useful. Later programs released tolerant elms such as Liberty, Valley Forge and New Horizon.
LessonA street planted with one kind of tree can lose all of it to one pest.
Credit
elm breeders at the University of Wisconsin, the US National Arboretum and the Morton Arboretum
1983
Wound paint fails the test
Alex Shigo of the USDA Forest Service showed that trees do not heal decay; they wall it off inside the wood, a model he called compartmentalization (CODIT). In 1983 Shigo and Walter Shortle reported 13 years of trials in New Hampshire and Maine in which tar, asphalt paints, shellac and other materials were put on drilled and chiseled wounds, and 275 treated trees were cut open. "No material prevented decay." The individual tree mattered more than the treatment.
LessonJudge a tree treatment by cutting the wood open years later, not by how the wound looks.
Credit
Alex L. Shigo and Walter C. Shortle, USDA Forest Service
2002
Emerald ash borer found
Beetles collected from dying ash trees near Detroit were identified on 9 July 2002 as Agrilus planipennis, a beetle from Asia whose larvae eat the living layer under the bark. By autumn 2004 about 15 million ash trees in southeastern Michigan were dead or dying, and ash made up an average of 12% of street trees in nine cities sampled there before the beetle was found (Poland and McCullough, Journal of Forestry, 2006).
LessonThe same authors note that many cities replaced dead elms with maple or ash, "which now are threatened" by the Asian longhorned beetle and the emerald ash borer.
Credit
Michigan State University entomologists and taxonomist Eduard Jendek, who identified it; Therese Poland (USDA Forest Service) and Deborah McCullough (Michigan State University), who documented it
2006
Free software values trees
The USDA Forest Service and partners released the i-Tree software suite in August 2006. It turns a tree count or a sample of plots into estimates of what the trees do, such as air pollution removed and stormwater held, so a city can put the work in numbers. It is free, and i-Tree says thousands of communities, consultants and students worldwide have used it.
Credit
USDA Forest Service, with Davey Tree, the Arbor Day Foundation, the International Society of Arboriculture, Casey Trees and the Urban and Community Forestry Society
2022
Street trees seen by computer
A Google team joined the public tree inventories of 23 North American cities with street-level and aerial photos to make the Auto Arborist dataset: more than 2.5 million trees of 344 genera, which its authors say is more than 100 times larger than the closest earlier dataset. They note that city tree inventories are made by hand, cost "millions of dollars per census", and so are done rarely. The paper was peer reviewed and published at CVPR 2022, a computer vision conference.
Credit
Sara Beery, Jonathan Huang and colleagues at Google
Watch for
Chainsaw drone trims branches
A University of Canterbury team built a drone that carries a small battery chainsaw to cut branches thicker than 30 mm near live power lines. The pilot taps a branch on a screen and AI software flies the cut; the team said it hoped to produce the drone commercially in 2026.
Achieved whena utility or tree care firm outside the university publishes a field trial with the number of branches cut, misses and incidents, or the drone is sold and used in routine work.
Claimed by
Richard Green's team, University of Canterbury, New Zealand, September 2025
Open
01
Score a defect finder
A phone tool that marks visible tree defects in photos, such as dead branches, cracks, cavities and fungus growing on the trunk, tested against certified arborists who inspected the same trees in person.
The briefCredit · unclaimed02
Grade pruning cuts from photos
A phone tool that looks at a photo of a pruning cut and says whether it is a proper cut just outside the branch collar (the swollen ring where a branch joins the trunk), a flush cut that removes the collar, or a stub left too long.
The briefCredit · unclaimed03
Robot makes a collar cut
A low-cost robot arm with a small motorized hand pruner that removes side branches from limbs clamped on a bench, placing each cut just outside the branch collar.
The briefCredit · unclaimed
Open 01 · the brief
Score a defect finder
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A phone tool that marks visible tree defects in photos, such as dead branches, cracks, cavities and fungus growing on the trunk, tested against certified arborists who inspected the same trees in person.
Why it is still open
Tree risk checks are still done by eye with a form, and arborists do not always agree: Klein, Koeser, McBride and Hauer (2023) found that arborists with ISA's Tree Risk Assessment Qualification (TRAQ) rated the same filmed trees as high or extreme risk less often than arborists without it. The closest image work, by Apostolov, Oke, Suttle, Arwade and Kane at UMass Amherst (2023), trained a model on 665 photos of trees along Eversource power lines in Massachusetts, each with one field rating of likelihood of failure; the team named disagreement among assessors as a limit and checking against several arborists as the next step. TH Wildau's TreeAngel prototype (2025) looked only for dead wood in 225 images. Several defect detectors on GitHub (2024 to 2026) report no comparison with arborists, and web, Crossref and GitHub searches found no public photo set of trees with defects marked by several qualified assessors.
Built with what exists
A phone camera; CVAT or Label Studio for marking defects; an open object-detection model such as Ultralytics YOLO, or an open vision-language model run locally with ggml (see the Note "A first stable number for local AI"); ISA's Basic Tree Risk Assessment Form as the list of things to label.
First step
Ask a city forester for the risk records of 20 street trees already inspected, photograph those trees, and mark their visible defects.
Done looks like
200 public trees, each photographed from four sides and inspected in person by three TRAQ-qualified arborists. The photos and forms are published, and the tool's hit rate and false alarms for each defect type are reported next to the arborists' agreement with each other.
Open 02 · the brief
Grade pruning cuts from photos
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A phone tool that looks at a photo of a pruning cut and says whether it is a proper cut just outside the branch collar (the swollen ring where a branch joins the trunk), a flush cut that removes the collar, or a stub left too long.
Why it is still open
Shigo's research showed that cut placement, not wound paint, decides how well a tree walls off decay, and researchers still test placement by cutting trees open: Miesbauer, Gilman, Koeser, Hauer and colleagues (preprint, April 2026) pruned 194 branches that had no collar at two angles and examined them three years later. A crew's cuts are checked by eye, if at all. A search found image work that picks pruning points on grapevines and fruit trees, such as Fernandes and colleagues (2025), whose vineyard system succeeded on 54.2% of test vines, but no tool that grades finished cuts on landscape trees and no public set of labeled cut photos; GitHub searches for pruning cuts and branch collars found none either.
Built with what exists
A phone camera; CVAT for labeling; a small image classifier fine-tuned in PyTorch, or an open vision-language model run locally with ggml (see the Note "A first stable number for local AI").
First step
With the grounds manager's permission, photograph 50 pruning cuts in a local park or campus and have one arborist label them.
Done looks like
500 photographed cuts on park or campus trees, each labeled by two ISA Certified Arborists as collar cut, flush cut or stub. On 100 held-out cuts, the tool's agreement with the arborists' consensus is published next to the two arborists' agreement with each other.
Open 03 · the brief
Robot makes a collar cut
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A low-cost robot arm with a small motorized hand pruner that removes side branches from limbs clamped on a bench, placing each cut just outside the branch collar.
Why it is still open
Pruning robots under study work in orchards and vineyards, where the goal is fruit, not the tree's long-term health: the USDA ARS RoTSE system measures fruit trees to decide which branches to cut, and the vineyard system above succeeded on 54.2% of vines. Tree-work robots aim at reach and safety: Canterbury's chainsaw drone clears branches from power lines, and a 2026 student robot climbs trunks with a chainsaw under remote control. Web and GitHub searches found no robot scored on where it places cuts on landscape trees.
Built with what exists
An SO-101 arm with LeRobot (see the Note "No download"); a control model trained from demonstrations with NVIDIA GR00T 2 (see the Note "Five demonstrations per task") or Runway Praxis-1 (see the Note "Open robot model weights"); a servo-driven bypass pruner for branches up to about 15 mm thick; a wrist camera; maple or oak limbs from yard waste held in a vise.
First step
Collect 30 small limbs with side branches from yard waste, clamp one in a vise, and record 50 remote-controlled cuts with the arm.
Done looks like
50 filmed cuts on limbs the robot has not seen, each scored by an ISA Certified Arborist as collar cut, flush cut or stub, with the published tally and seconds per cut against a person with hand pruners.