Industries / Construction and property / Commercial and residential construction / Concrete
Finishing
Finishing is the work done on the top of a fresh slab as it stiffens: striking it off to height (screeding), floating and troweling it smooth and dense, and checking how flat and level it came out.
5
Milestones achieved, from 1940 to 2019.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1940
A machine floats and trowels
Done by hand, finishing a floor takes many separate passes. Eugene C. McMillan's US patent 2,208,801, filed in March 1939 and published in July 1940, describes a walk-behind electric machine with rotating rings that both floats and trowels a floor, to "eliminate the hard manual labor which is now involved in cement finishing". It says powered floating machines had by then "been used for many years". By 1996, Japanese researchers wrote, walk-behind and ride-on power trowels were the machines most often used to finish floors on sites in Japan.
Credit
Eugene C. McMillan
1986
Lasers set the screed
On a large slab, crews pull screeds over the fresh concrete and correct its height by eye and hand. A laser screed holds its striking head to a plane set by a laser, adjusting it continuously as it works. Somero Enterprises, of the US, says it started the laser screed market in 1986, and that its machines, used on warehouse and factory floors, make floors "faster, flatter and with fewer people", its own claim.
Credit
Somero Enterprises
1987
Flatness gets a number
Slab finish was specified by the gap under a 10 ft (3 m) straightedge, with no national standard for where to set it or how many readings to take, which led to conflict and lawsuits, writes Mark Cheek in the American Concrete Institute's Concrete International. Allen Face's F-number system, published in Construction Specifier in April 1987, scores flatness (FF) from the curvature of elevation readings over 24 in. (600 mm) and levelness (FL) from slope over 10 ft (3 m), and ASTM E1155 sets how the readings are taken across a floor.
Credit
Allen Face; ASTM; ACI
1996
Robots stop at edges
The number of plasterers in Japan, who finish floors there, fell from about 230,000 at its peak to about 50,000, and several floor troweling robots were built in response, but they had not spread: they were heavy, took long setup and cost too much, by the account of Masaaki Kobayashi of the builder Hazama and co-authors from Mitsubishi Corp. and Eroica. Their 73 kg robot finished 1,000 m² on five sites about as smoothly as plasterers, but it left tire marks when it turned sharply and could not work near columns and forms.
LessonA robot that cannot reach the edges still needs the finisher it was built to replace.
Credit
Masaaki Kobayashi, Masaru Soma and Takanori Shigihara
2019
Scanners fail the flatness test
ASTM E1155 allows laser imaging devices, such as 3D laser scanners, to collect F-number data, but its precision statement covers only profilers of the Dipstick's type. In October 2018 the American Society of Concrete Contractors (ASCC) had eight teams scan slabs at a site in Walnut Creek, California, and compared their F-numbers with those from a Dipstick, a walking profiler. The scanners repeated their own results about as well as the profiler, but their flatness numbers came out consistently higher, for a reason the authors called unclear, and in October 2019 ASCC said it did not recommend laser imaging devices for specification compliance. A second study, in which 13 participants scanned a slab in Santa Cruz, California, published in ACI's Concrete International in January 2025, found smaller errors but supported scanning only for tolerances of ½ in (13 mm) or more, which ASCC reads as ruling scanners out for F-numbers.
LessonA new measuring tool needs published repeatability figures before contracts can rely on it.
Credit
ASCC; William Paul, James Klinger and Bruce Suprenant; Leo Zhang and co-authors
Watch for
Floors measured for robots
Floor Agent, a floor construction company in Tokyo, says warehouse robots that carry goods (AGVs and AMRs) need precise concrete floors, and showed a "Floor Profiler" that it says measures floor level in steps of 0.1 mm, along with construction robots and a carbon composite product that smooths the steps at joints.
Achieved whenan outside party publishes the Floor Profiler's readings beside those of a Type II profiler under ASTM E1155 on the same lines, with repeatability.
Claimed by
Floor Agent, November 2026
Open
01
Robot finishing at the edges
A robot arm that finishes by hand-tool the strips that troweling machines leave: the band next to a form edge and the ring around a pipe or column, on small test panels of fresh concrete, starting when a penetration test says the surface is ready.
The briefCredit · unclaimed02
Profiler mistakes caught early
An open program that computes FF and FL from the raw readings of a walking floor profiler by the ASTM E1155 method, and flags the patterns that come from testing mistakes rather than from the floor: a sharp step where a reading was taken before the foot touched down, a spike where a foot rode over a pebble or dust, and a sawtooth from one miscalibrated side. Test it on runs with mistakes made on purpose.
The briefCredit · unclaimed
Open 01 · the brief
Robot finishing at the edges
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A robot arm that finishes by hand-tool the strips that troweling machines leave: the band next to a form edge and the ring around a pipe or column, on small test panels of fresh concrete, starting when a penetration test says the surface is ready.
Why it is still open
The 1996 Hazama robot could not work near columns and forms, and its designers wrote that finishers would still be needed there. Jia-Rui Lin and colleagues' 2025 paper on swing-arm troweling robots plans which floor areas such a robot can reach around obstacles, tested on obstacle maps, not on concrete. No published test has a robot finish the edges themselves.
Built with what exists
A collaborative arm with force sensing, such as a Universal Robots UR10e or a Franka Research 3; a steel finishing trowel and an edger in a 3D-printed holder; a wrist camera; Hugging Face's LeRobot to record demonstrations and train; an open robot model such as Runway's Praxis-1 (see the Note "Open robot model weights") or NVIDIA's GR00T 2 (see the Note "Five demonstrations per task"); a penetration resistance needle of the kind used in ASTM C403 to time the start; a 1 m straightedge and feeler gauges.
First step
Cast five panels in plywood forms, have a finisher work the edges while the arm records in hand-guiding mode, and take penetration readings every 15 minutes to find the working window.
Done looks like
Twenty filmed panels in a row, each with a 1 m edge strip and one pipe through it, finished by the arm alone; the largest gap under a 1 m straightedge measured on each; and the panels graded blind by two finishers beside 20 panels finished by hand, with all recordings and scores published.
Open 02 · the brief
Profiler mistakes caught early
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open program that computes FF and FL from the raw readings of a walking floor profiler by the ASTM E1155 method, and flags the patterns that come from testing mistakes rather than from the floor: a sharp step where a reading was taken before the foot touched down, a spike where a foot rode over a pebble or dust, and a sawtooth from one miscalibrated side. Test it on runs with mistakes made on purpose.
Why it is still open
Profilers remain the accepted tool for F-numbers (see 2019 above), and their results are taken on trust. ASCC's technical director wrote in November 2025, drawing on Bill Arpin's account, that a single false reading can turn an FF 65 floor into FF 28, that a 1/16 in (1.6 mm) calibration offset can fail a floor, that few people question the raw data behind a report, and that owners should ask for the profiler plots to spot such errors by eye. A search found no profiler software that flags them; Leica's point-cloud FF/FL tool, Cyclone 3DR, warns about sample lines that are too few, too short or over sparse data, not about bad readings, and the open floor-flatness programs posted on GitHub in 2026 work from scans and do not compute E1155 F-numbers.
Built with what exists
A rented walking profiler such as a Dipstick or F-Meter, which records elevations at 12 in (300 mm) steps; Python with NumPy for the E1155 calculation; simple rules plus a small classifier trained on labeled runs to find steps, spikes and sawtooth patterns; washers, a pinch of grit and shims to make the mistakes.
First step
Write the E1155 calculation in Python and check it against the maker's software on ten runs from one floor.
Done looks like
60 profiler runs on finished floors, 30 of them with one planted mistake of a recorded kind and position, published with the code; the program flags at least 27 of the 30 and no more than 3 of the 30 clean runs, and its F-numbers on clean runs match the profiler maker's software within 0.5.