Industries / Consumer goods / Furniture
Cutting and machining wood
Turning boards and panels into furniture parts by sawing, routing, drilling and gluing strips over raw edges, with machines fed by hand or run by computer.
5
Milestones achieved, from 1870s to 2025.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1870s
Steam moves the factories
Grand Rapids, Michigan, grew as a furniture town on the Grand River, whose 18-foot (5.5 m) fall powered its first mills and woodworking shops. When steam engines replaced water power in the 1860s and 1870s, factories moved away from the river to sit beside railroad tracks, and brick mills with tall smokestacks spread across the city. After Grand Rapids furniture was shown at the Philadelphia Centennial Exhibition in 1876, the city became known as the furniture capital of America, and buyers came to its furniture markets, first held in 1878.
Credit
Grand Rapids makers such as the John Widdicomb Company, founded in 1865
1962
Edges glued by machine
Edgebanding glues a thin strip over the raw edge of a panel such as particleboard. In 1962, two years after it was founded in Germany, HOMAG patented the KH 2/18, which it calls the first automatic edge banding machine. The same year it showed a machine using hot-melt glue that bonded edges faster and more cheaply than its rivals. In 2009 it showed laserTec, which melts the back of the edge strip with a laser beam and presses it on, leaving almost no visible joint. HOMAG says it now holds more than 30% of this market (its own figure).
Credit
HOMAG
1968
Furniture dust and nasal cancer
E. D. Acheson and colleagues reported nasal cancer among woodworkers in the furniture industry in the British Medical Journal in 1968. A later follow-up of 5,371 men who had worked in nine furniture factories in Buckinghamshire, England, found adenocarcinoma of the nose, a cancer of the gland cells lining it, at about 100 times the rate expected in the local population. A 1984 study of 5,108 of these men found no rise in deaths from other cancers.
LessonWithin the factories the cancer was more common the dustier the job, which pointed to the wood dust itself (International Journal of Epidemiology, 1981).
Credit
E. D. Acheson, R. H. Cowdell, E. Hadfield and R. G. Macbeth
2020
Robots drill kitchens to order
At its plant in Sélestat, France, the kitchen maker Schmidt runs a cell of five ABB industrial robots, built by Zimmer Group, that drills, grooves and fits hardware to each cabinet part as it is ordered, a way of working the trade calls "batch size 1". Parts follow each other less than two seconds apart with an accuracy of 0.2 mm, and a side panel 2 m tall can be followed by a 15 × 20 cm shelf with no time spent changing the setup. Software writes each part's machining program from the order database.
LessonRobot arms were not accurate enough for this at first, and Zimmer had to measure each robot with a laser tracker and correct for its errors in software (Achim Gauß, Zimmer Group).
Credit
Schmidt Groupe and Zimmer Group
2025
Saw safety rule withdrawn
Table saws are the main cutting tool in many small furniture and cabinet shops. In 2003 Stephen Gass and others connected with SawStop, a maker of saws that sense contact with a person and act to prevent or limit the injury, asked the CPSC to require such protection. The CPSC proposed in 2017, and again in 2023, that a saw cut no deeper than 3.5 mm into a test probe moving toward the blade at 1 m/s. It estimated 26,500 emergency-room injuries from blade contact in 2017, including 2,800 amputations, and said the rule would prevent or lessen 49,176 injuries a year. On 29 September 2025 the CPSC withdrew the proposal and said it did not intend to issue a final rule.
LessonAfter 22 years of rulemaking, a safeguard already on sale stayed optional in the United States.
Credit
the CPSC; the 2003 petition by Gass and colleagues
Watch for
Robot cells that run unattended
At Ligna 2025 HOMAG showed a nesting cell, a CNC router that cuts many parts from one sheet, whose STACKBOT C-300 robot uses a camera to recognize the cut parts without barcodes and sort them onto up to eight stacks. It also showed a DRILLTEQ V-310 drilling cell with robot loading that it says gives "continuous 24/7 production" while handling parts one at a time in mixed order.
Achieved whena furniture plant using one publishes, or an outside party measures, its unattended running hours and its rate of wrongly identified parts over several months.
Claimed by
HOMAG, May 2025
Open
01
Public cut lists from shops
An open benchmark of real cabinet and furniture cutting jobs. Each job gives the part list (sizes, quantities, material, grain direction, which edges get banding), the sheet sizes in stock and the layout the shop actually cut, with customer details removed. Then a fair race of cutting optimizers, open and commercial, on those jobs, scoring sheets used, offcuts worth keeping and saw cycles.
The briefCredit · unclaimed02
Edge checks that travel
A camera check for edgebanding faults, such as gaps, glue lines, chipping and strips that overhang, that is trained at some shops and still works at a shop it has never seen. The core is a test set from at least ten shops with different machines, edge materials and lighting, with each image labeled by two trained inspectors.
The briefCredit · unclaimed
Open 01 · the brief
Public cut lists from shops
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open benchmark of real cabinet and furniture cutting jobs. Each job gives the part list (sizes, quantities, material, grain direction, which edges get banding), the sheet sizes in stock and the layout the shop actually cut, with customer details removed. Then a fair race of cutting optimizers, open and commercial, on those jobs, scoring sheets used, offcuts worth keeping and saw cycles.
Why it is still open
Shops choose among commercial optimizers and dozens of free projects on GitHub, with no shared test on real jobs. The 2DPackLib library gathers academic benchmark sets for two-dimensional cutting in one format, but they come from the research literature, not from furniture shops' orders. Vanzela and colleagues (Computers & Operations Research, 2017) modeled cutting with saw-cycle limits for a furniture factory, but no public copy of its job data was found. Searches in November 2026 found no public set of real furniture cutting jobs.
Built with what exists
Python; Google OR-Tools' CP-SAT solver for a strong baseline; open guillotine-cutting projects on GitHub; trial versions of commercial optimizers; the 2DPackLib file format, so results can be compared with the academic sets.
First step
Ask three local cabinet shops to export last month's cut lists from their design software, remove customer names, and convert them to the 2DPackLib format.
Done looks like
200 anonymized jobs from at least five shops released under an open license, with each optimizer's sheet count, waste percentage and saw cycles for every job, and a second team rerunning the scoring and getting the same numbers.
Open 02 · the brief
Edge checks that travel
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A camera check for edgebanding faults, such as gaps, glue lines, chipping and strips that overhang, that is trained at some shops and still works at a shop it has never seen. The core is a test set from at least ten shops with different machines, edge materials and lighting, with each image labeled by two trained inspectors.
Why it is still open
Two published detectors were each trained on one company's images. Fu and colleagues (Nanjing Forestry University, 2025) used 1,887 images of six defect types from four production lines, reported 82.1% precision and 75.4% recall on camera images from the line, and named the narrow data and untested lighting changes as limits. Xiao and colleagues (PLOS ONE, 2026) released 6,000 images in eight defect classes from one factory, Xiamen GoldenHome, labeled by one person, and reached 74.0% mean average precision. Neither tested a model at another plant.
Built with what exists
A phone or industrial camera with fixed lighting; Label Studio for labeling; the open Zenodo set from Xiao and colleagues for pretraining; an Ultralytics YOLO11 detector.
First step
Photograph 200 edged panels at one cabinet shop under its normal lights and ask two people who check edges there to label them separately.
Done looks like
3,000 labeled images from ten shops, published under an open license. For each shop, a model trained without that shop's images is tested on them, and its recall for each defect type is reported next to how often the two inspectors agree with each other.