Matteo Campagna
ITEN
Menu
Menu

Method

Shooting machines and robots: testing the equipment without the archer

Since 1990 engineers have mounted the bow on a frame or a robot to take the archer out of the experiment and measure the equipment alone. This page explains what has been measured that way, how tightly a machine groups, where its release differs from the fingers and which traps lie in the comparisons with archers.

Studies in the registry
21
On archers
0
Years
1990-2025
First study in the registry
1990
Median sample size
31

Why take the archer out

When an arrow leaves the group, the target face does not say whether the archer or the equipment was responsible, and a shooting machine exists to pull the two apart. It is a frame that holds the bow the same way every time and lets go of the string the same way every time, so that whatever differences remain on the target belong to the equipment, to the air or to the machine itself. The first example in the registry is the Polish study of 1990: a very stiff frame set in concrete, whose arrows at 30 m had a standard deviation of 0.9 cm. With it the author measured something no archer could have isolated, how far the impact moves for one degree of bow torque: 15.8 cm in one direction and 21.9 in the other.

Since then the machines have split into three families. There are frames that hold the bow and release the string with a mechanism, from 1990 to the commercial machines used on compound bows; robots with articulated fingers, built in Japan to reproduce the recurve release; and compressed-air launchers, which push the arrow without a bow and therefore without any flex. In 2022 they were joined by a humanoid robot that holds the bow and aims with a camera.

What has been measured this way

Most studies use the machine to compare equipment too similar for an archer to tell apart. A Taiwanese group compared vanes and strings: between the bare shaft and the slowest vane there were about 3.2 feet per second, or 1.5%, with standard deviations of 0.2 to 0.33. A group led by Park, with a compound on a machine bolted to the ground, saw the group at 25 m grow from 27 to 48 mm with shafts bent by six thousandths of an inch, and shrink from 44 to 27 mm when only the nocks were changed. The same machines have measured wind drift: at 70 m, in gusts above 10 m/s, two arrow models had lateral standard deviations of 44.6 and 58.6 mm, a ratio of 1.41 against the 1.45 the drift calculation predicted. The rest is in the line on wind and heat.

Machines are now also used to choose arrows. A 114 kg Japanese machine shot six arrows at 70 m and left the two defective ones outside the group, while a 2025 study of 59 X10s of the same model found that mass and mean spine explain 46.7% of the vertical position of the impacts, and that mean spine and the variation of spine around the shaft explain 18.6% of the horizontal one. The entry on the machine that picks out defective arrows stays at "reported", because nobody has checked that the arrows the machine rejects are the ones that lose points in competition.

How closely the machine imitates the archer

A machine is only useful if the arrow behaves as it does when a person shoots it, and here the evidence has split. In 1990 footage at a thousand frames per second showed no difference between the machine and a Polish national-team archer, but the comparison was made by eye. In 2014 a Japanese group launched the same arrows in four ways: the flex frequency stayed at around 86 Hz, while the amplitude was 6.86 mm with two top national archers, 14.8 with the robot's mechanical fingers and 1.22 with a mechanical release. Those figures underpin the entry on the release aid that makes the arrow flex less, and the same experiment found air drag more consistent in the archers' shots than in any machine's.

A year later the robot's fingers were redesigned: opened up to the smallest angle at which the string slips off, they produced 6.94 mm of flex against 6.86 for an international-level archer, with a standard deviation of 0.18 to 0.23 mm against 0.34. The Japanese machine of 2020 added two details taken from the shooting line: the grip lets the bow jump forward as it does in the hand, and the release is triggered by the sound of the clicker, so the draw length is always the same. This is why the entry on the machine that reproduces the archer's arrow says it can do so if it is built to; no comparison has involved more than two archers. A study done with a mechanical release on a recurve is therefore looking at an arrow that bends far less than one leaving the fingers, and the line on the archer's paradox explains why that matters.

How tightly a machine groups, and against whom

Well-built machines group very tightly: twelve arrows in a row through the same hole at 12 m, 13 mm at 25 m, ten arrows out of ten in the 122 mm ten ring at 70 m outdoors with no wind, and a median distance of 1.3 cm from the group centre at 70 m indoors. Only one study compares them on the target with an archer in the same facility: at 55 m the vertical standard deviation of the impacts was 107 and 108 mm for one archer, 34 for the robot and 9 for its improved version. But there was one archer, of unstated level, with different arrows from the robot's. This is why the entry on the machine that groups more tightly than the best archer stays at "reported": the tests on the machine are many and consistent, and those on the best archer are missing.

On launch speed the gap narrows. In seven elite German archers mean arrow speed varied by between 0.23 and 0.58% from shot to shot; in the laboratory an expert archer repeated her launch speed within 0.6%, against 0.3% for the compressed-air launcher, and a compound on a machine measured with the NIST reference chronograph varied by about 0.5%. So the expert archer is in the same order of magnitude as the machine, as the entry on shot-to-shot speed says, and the machine's advantage lies elsewhere in the shot.

The traps in the comparisons

The first trap is the machine itself: the Japanese one, at 24 kg in its first version, vibrated as the string was drawn and lost accuracy. A prototype with a training bow kept the same arrow within a 160 mm circle at 18 m, the target the authors had set themselves, and they put the remaining spread down to a nocking point wider than the nock, to the crosswind and to the lack of a clicker. The 2022 humanoid robot put ten arrows into 6.6 by 4.7 cm at 10 m, but with the group centre 27.85 and 25.96 cm from the centre of the target; it is a preprint, and it shows that a tight group can sit a long way from the middle.

The second is how the target is read. In the 2020 test, the standard deviation that separates the sound arrows from the defective ones was calculated on the same four sound arrows it then judges, so the fact that the sound ones stay inside the circle says little. In the index of a 2014 Korean study, three arrows in one hole and three arrows arranged in an equilateral triangle get the same score, zero: the index measures how regular the group is, not how tight.

The third is the opposite question, which a machine cannot ask: what happens when the bow changes and the archer does not. In 1990 eight archers with competition experience shot with their own bow and with an assigned one and showed no significant difference in precision, over nine arrows per distance. The entry on shooting a bow other than your own explains why a result based on so few shots leaves the question open.

How we came to know it

  1. 1990OriginPekalski 1990 Experimental and theoretical research in archeryA frame set in concrete releases the string in place of the archer: at 30 m its spread has a standard deviation of 0.9 cm, and it shows that one degree of bow torque moves the impact by 15.8 or 21.9 cm depending on the direction. Footage shows the arrow on the string looking the same with the machine and with a Polish national-team archer, though the comparison is made by eye.
  2. 2009ExtensionLin et al. 2009 Performance analysis based on an emulated archery machineA machine built on a single-axis robot compares vanes and strings: between the bare shaft and the slowest vane there are about 3.2 feet per second, or 1.5%, with standard deviations of 0.2 to 0.33. No archers in the study.
  3. 2011MethodPark 2011 The Behaviour of An Arrow Shot from a Compound Archery BowA compound on a machine photographed with timed flash every two milliseconds: the arrow's flex and lateral position repeat from shot to shot within the camera's resolution of about 0.2 mm, so they can be measured as a single curve.
  4. 2011ExtensionPark 2011 Minimizing wind drift of an arrowTwelve arrows in a row through the same hole at 12 m. At 70 m, in gusts above 10 m/s, the machine separates two arrow models with lateral standard deviations of 44.6 and 58.6 mm, a ratio of 1.41 against the 1.45 calculated for the drift.
  5. 2011MethodBarton et al. 2011 A miniaturised arrow ballistic measurement systemThe machine serves as the benchmark for an electronic point: four compound shots measured at the NIST range with the reference chronograph vary by 0.45 m/s, about 0.5%.
  6. 2012ExtensionMiyazaki et al. 2012 Aerodynamic properties of an archery arrowA compressed-air launcher repeats its launch speed within 0.3% up to 62 m/s; an expert Japanese national-team archer, in the same study, within 0.6% at 58 m/s.
  7. 2012ExtensionYu et al. 2012 Measuring System for Impact Point of Arrow using Mamdani Fuzzy Inference SystemA group of Korean engineers proposes testing each arrow at the factory by shooting it three times from a machine, turning the nock 120° each time, and reading its quality from the spread of the three impacts.
  8. 2014ContradictionAndo et al. 2014 J0520102 Aerodynamic properties of an archery arrow : Influence of flexural oscillationThe same arrows launched in four ways: the flex frequency stays at around 86 Hz, but the amplitude is 6.86 mm with two top national archers, 14.8 with the robot's mechanical fingers and 1.22 with the mechanical release. Air drag varies least in the archers' shots.
  9. 2014MethodJeong et al. 2014 A Study of Arrow Performance using Artificial Neural NetworkSeventy-nine arrows, with straightness and spine as the inputs of a neural network. The quality index, calculated on three impacts at 25 m, is zero both for three arrows in one hole and for three arrows in an equilateral triangle.
  10. 2015ExtensionWakatabe et al. 2015 2A2-T08 Development of Archery Robot : Development of finger mechanism for controllable the bending vibrationThe robot's fingers are redesigned: opened to the smallest angle at which the string slips off, they give 6.94 mm of flex against 6.86 for an international-level archer, with a standard deviation of 0.18 to 0.23 mm against 0.34.
  11. 2017ExtensionPark et al. 2017 Effect of arrow shaft straightness on arrow groupingA compound on a machine bolted to the ground, groups at 25 m: shafts bent by six thousandths of an inch widen the group from 27 to 48 mm, changing only the nocks takes it from 44 to 27, and the tightest measures 13 mm.
  12. 2019ExtensionNakamura et al. 2019 Statics Investigation about Archery Robot in Drawing a BowA prototype robot with a training bow shoots the same arrow ten times at 18 m and keeps it within a 160 mm circle, the target the authors had set themselves with that bow. They put the remaining spread down to a nocking point wider than the nock, to the crosswind and to the lack of a clicker: how tightly a machine groups depends on how it is built.
  13. 2020ExtensionOhara et al. 2020 Development of an Archery Robot for the Selection of ArrowsThe second version of a Japanese machine goes from 24 to 114 kg and releases at the sound of the clicker: ten arrows out of ten in the ten ring at 70 m, and two defective arrows outside the group of four sound ones, with the standard deviation calculated on those four.
  14. 2022ExtensionChristmann et al. 2022 Can a Robot Shoot an Olympic Recurve Bow? A preliminary studyA humanoid robot draws a 16 lb recurve and aims with a camera: at 10 m the ten impacts fall within 6.6 by 4.7 cm, with the group centre 27.85 and 25.96 cm from the centre of the target. It is a preprint.
  15. 2022ExtensionMurakami et al. 2022 Automatic image analysis of an archery arrow in free flight using high-speed video camerasThe only comparison on the target in the same facility, at 55 m: the vertical standard deviation of the impacts is 107 to 108 mm for one archer, 34 for the robot and 9 for its improved version. There is a single archer, of unstated level, shooting different arrows from the robot's.
  16. 2025ExtensionKuch et al. 2025 Improving the arrow selection process in archery: the stiffness variation pattern mattersFifty-nine X10s of the same model shot by a machine at 70 m indoors: the same arrow lands at a median distance of 1.3 cm from the group centre, and mass and mean spine explain 46.7% of the vertical position of the impacts.

Studies that use it

Related claims

Claims in the registry with at least one of these studies among their evidence.

66 claims · 23% with evidence on archers

Articles

Research lines

Cite
APA

Campagna, M. (2026). Shooting machines and robots in archery research. In Research. matteocampagna.com. Retrieved 16 October 2026, from https://matteocampagna.com/en/research/methods/shooting-machines-robots

BibTeX
@misc{campagna2026_enresearchmethodsshootingmachinesrobots,
  author = {Campagna, Matteo},
  title = {Shooting machines and robots in archery research},
  year = {2026},
  howpublished = {Research, matteocampagna.com},
  url = {https://matteocampagna.com/en/research/methods/shooting-machines-robots},
  note = {accessed 2026-10-16}
}