There is no shortage of predictions about the future of archery: headsets for training at home, sensors that correct your technique, brains trained with feedback. Some of these promises, though, have already been tested on archers, with published numbers, and the measurements say more precise things than any forecast. Some technologies have already done well on narrow tasks: a program that reads competition video times the interval between anchor and release with a mean error of 0.34 seconds, and in a test with six arrows a machine shooting at 70 metres picked out the two defective ones from the sound ones. Others have shown their limits: in a test on twenty-two beginners, scores were higher in virtual reality than with a bow, but the body worked less hard, and the only controlled trial of neurofeedback in archers saw the trained signal change, not the score.
In this article I set out what has been measured, with the year it was measured, and only at the end do I say what I take from it for the years ahead. Each claim links to its entry in the registry, which shows how many studies it rests on and what status it has.
Measuring the shot outside the lab
For decades the biomechanics of the shot was studied in the laboratory, with multi-camera optical systems and markers on the body. The first question, then, is whether simpler instruments give comparable numbers. For wearable inertial sensors there is a single direct comparison, a 2024 conference abstract: six archers from the Chinese national team shoot at fifty metres wearing five sensors while an eight-camera optical system records them. In most cases the acceleration curves from the two systems correlate above 0.75, meaning the sensors follow the shape of the movement as the optical system does. A high correlation does not guarantee that the two instruments give the same number, though: a sensor can reproduce the shape of a curve well and still misjudge its size, and the abstract does not report the errors that would tell us.
Simpler still is a wrist accelerometer on the bow hand. In a 2025 study released as a preprint, six teenage archers wore one in an official competition, and a neural network had to pick out the stretches of signal containing a shot, from the start of the draw to the release. Given only the three raw axes it manages this in 84.5% of cases, and with two extra quantities calculated from the signal in 96.0%. The sensor is the same in both cases, so the gain comes from processing the signal. The model, however, was not tested on archers it had never seen, and the measure the authors propose for real use only rises from 67.9% to 75%.
Video is the tool almost every archer already has, and here there are more tests. The strongest result dates from 2024: on fifty-eight competition clips of world-class archers, filmed at thirty frames per second, a system based on pose estimation times the interval between anchor and release with a mean error of 0.34 seconds against an expert's manual measurement. It works because at anchor the angle of the drawing arm almost stops changing; if the program watches the string near the face instead, the error rises to 1.4 seconds. In 2025 a classifier built on joint positions extracted from the videos of twenty archers recognised three phases of the shot with 90% accuracy in recurve and 76% in barebow, and it fails exactly where posture changes least, between draw-and-aim and the expansion into the release. An error of a third of a second is enough to log the timings of a competition, not to judge a single arrow. For using video in your own training, see the article on video analysis.
How well a pose estimation model locates an archer's joints in ordinary footage, on the other hand, has not yet been measured in a way that can be read. The only check of the error, from 2026, uses eight photos of a single archer, one per phase of the shot, and reports a normalised error between 0.55 and 0.71 without defining exactly the scale that would tell us whether that is small or large: that is why the entry is Not studied. That the tracking carries information shows up in another paper. In 2025, using only the trajectories of the shoulders, elbows and wrists extracted from 5,984 shots, a neural network identified which of four Korean archers was shooting in 98.7% of test cases. It is a sign that each archer's movement carries a recognisable signature, even though the choice was between only four people and another network tried on the same data stopped at 55.9%.
The brain and the eyes
Neuroscience is where the biggest promises are made, and also where the numbers need reading most carefully. On EEG the best data date from 2022: fourteen archers from the Chinese national team and fourteen from a provincial squad shoot thirty arrows wired to an electroencephalogram, and in the last second of aiming the national archers' brains work in a more coordinated way, with a more integrated and efficient network. The picture rests in practice on a single research group, though, and a comparison between groups that already differ does not say whether training is what gets the brain working that way.
Training that signal with neurofeedback has been tested on archers, with a control group and original data that can be read, in a single study from 2012, and the result goes against the promise. Twenty-four Indian university archers, twelve with twelve sessions over four weeks and twelve without, shot a 72-arrow competition at 70 metres before and after: the trained group went from 527.3 to 529.9 points, the control group from 526.6 to 526.7, and there is no difference between the groups (p = 0.561). The trained signal moved and the score did not, which is why the entry is Contradicted.
For gaze, the step forward has been in measurement. The quiet eye, the final fixation on the target before the release, is usually measured with a system that films the eye. In 2025 a group at the University of Mersin derived it from electro-oculography instead, the electrical potential recorded around the eye: in ten club archers, six arrows each, neural networks trained on an expert's annotations recognise the quiet eye with about 93% accuracy and, with the cleaned signal, miss its onset by 7 to 9 milliseconds. On the raw trace the best network drops to 84%, and the reference is still one expert's judgement rather than the video measure the quiet eye literature relies on.
That leaves computer-based perceptual training, which is often put forward as a way to train concentration. In archers there is a single experiment, reported in two papers from 2020 and 2021: twenty archers from an Indonesian university club, ten with twelve NeuroTracker sessions over four weeks, tracking moving targets with their eyes on a 3D television, and ten with their usual training. At 30 metres the first group gains 32.9 points and the second 19.7. However, the control group also improves a great deal and does not spend an equivalent time at a screen on a dummy task, and the paper on scores appeared in a journal with questionable editorial standards: the registry reports the result without using it to decide the status, and the entry is Not studied.
Virtual reality
There is only one comparison between headset and bow, from 2024, and it involves twenty-two university students at the end of a fourteen-week archery course they began with no experience. Each shot ten arrows at eighteen metres with a bow and ten in virtual reality, in counterbalanced order. In the headset they scored more, 86.9 against 78.8, but the body worked less: with the bow heart rate rose 32.5% above rest, in the headset 17.5%, and real shooting was rated harder and more tiring, while reported anxiety was identical. The authors read the higher score as the sign of an easier simulation, not of better shooting.
Two details weigh on the reading. The bow test was the course's final exam, while the headset test did not count; and for someone who has been shooting for a few months, drawing a bow is still a real physical effort, which a controller with no draw weight does not ask for. So there is no telling how much of the difference comes from muscular load, how much from the exam and how much from virtual shooting itself. The registry counts those students as the general population, so the entry on arousal is Transferred, and nobody has checked whether training in a headset moves the score with a bow: until that is known, simulator scores cannot be set beside range scores.
Machines, cameras and equipment
A shooting machine takes the archer out of the equation, so that arrows and bows can be tested without the variability of a human shot, and the first question is whether it shoots like an archer. The first comparison, in 1990, filmed a member of the Polish national team and a machine with the same bow at a thousand frames per second and saw no difference in the arrow's motion on the string, but the judgement was only qualitative. In 2014 the measurement became precise: the arrow vibrates in flight at around 86 Hz whoever launches it, but with a robot's mechanical fingers it flexes 2.2 times as much as with archers, and the arrow's drag is more consistent when an archer shoots it. A year later the same group redesigned the fingers and brought the flex down to 6.94 millimetres, against 6.86 for an international-level archer. Drag has not been measured again with the new robot, though, and no comparison involves more than two archers: that is why the entry is Contested.
For sorting arrows, in the tests so far, the machine is consistent enough. In 2020 a 114 kg machine with a recurve bow shot six arrows at 70 metres on a day with little wind: the four sound ones stayed inside a circle with a radius of two standard deviations, while the two defective ones landed outside it, each always on the same side, the one with a missing vane low and the other to the right. In 2025, fifty-nine Easton X10s of the same model, good arrows that differ only slightly, shot at 70 metres indoors, show how repeatable the system is: the same arrow shot thirty times lands a median 1.3 centimetres from the group centre, and mass and mean spine account for 46.7% of the arrows' vertical position. For comparison, at 55 metres an archer spreads 107-108 millimetres vertically, against 34 for a robot and 9 for its improved version. Whether the arrows a machine rejects are the ones that lose points in competition, however, nobody has checked.
Automatic scoring with a camera has also been in testing for years, and every prototype misreads some arrows. A 2017 thesis, written at the request of the Dutch federation, gave 29 arrows out of 32 the correct zone in six training videos, and the three misses had three different causes: one arrow shifted the target face as it went in and the program read a ten as 9.85; another stuck out too little in the image to be recognised as a shaft; a third arrived during a sharp change in light and was discarded. In 2025 a system tested in the field on a hundred shots by ten archers, against the judges' manual scoring, failed to see 9 arrows and gave the correct score to 67. Tests on arrows read one by one go no further than a hundred arrows, and none has yet been run in competition.
The last example is the furthest from everyday use. In 2013 a 34-gram magnetic sensor mounted on the bow was used as an audio sight: it turns the bow's direction into sound in headphones, with the side of the target in the right or left ear and up and down carried by two tones. Six trained, sighted university archers, blindfolded, scored at 10 metres, with a 16 lb bow, 58% of what they scored with their eyes open. So it is possible to shoot by ear and hit the target, but no blind archer was tested, and the study had no blindfolded trial without the sound, which would show how much the device really helps.
What I take from it for the years ahead
Set side by side, the measurements show where technology is already useful and where it promises more than it has shown. The best results concern timing and recognition: timing anchor and release from video, picking out the shots in a wristband's signal, sorting arrows with a machine. Camera scoring, by contrast, still misreads some arrows in every prototype tested. Fine measurements are still weak, such as joint angles from ordinary footage or the size of a movement from a sensor, because their error has not been measured in a way that can be read. And the promises to raise the score, from neurofeedback to computer training, do not yet have a test the registry can count in their favour.
The same goes for models meant to predict the score from data. Experience and past results carry a lot of weight, but in thirty-two Indian archers the second most important predictor was resting systolic blood pressure, a measurement taken on the day, and the best model explained little more than half of the variability: the entry is Contested, and nobody has yet compared a model with the simplest forecast of all, the average of the last few competitions.
The idea that technical knowledge is now within everyone's reach thanks to the internet also has a measurement, a small one. In a 2023 British thesis, nearly all of 63 qualified coaches used the internet, 90.3%, but only 21.9% consulted online resources every week, and in interviews they said they learnt mainly from other coaches, in person, and used the web to find things they already knew. The most experienced also used it to fill gaps, looking at other sports, which is why the entry is Contested.
My forecast, and it is a forecast rather than data, is that over the next few years the tools that time the shot and recognise its phases will enter training first, because they already make few errors on a simple task, and the tools that claim to correct technique arrow by arrow will come later. In the meantime, every new tool deserves the question the studies above asked: how large its error is against a reference, and on how many archers it has been tested.
What stays the same
The tools measured so far serve to observe the shot, and the shooting machine to test equipment: none of them does the archer's work. Stance, anchor, draw and release depend on anatomy and on the bow, and the most advanced systems serve to measure them better, not to change them. The same applies to the mental side. In statistical terms a good share of the differences in score between archers goes together with mental factors: in the largest study, from 1986, the mental block accounted for 26% of the differences, and in 61 athletes from the Chinese national team psychological factors explain about a third of the variability in results. These are associations, and they do not say how many points changing them would gain, but they are enough to show that the work of mental training still belongs to the archer and the coach, whatever tool measures it.
The risks
The risk I see most often on the range is dependence on the tool: an archer who gets used to a number after every arrow can stop listening to their own shot. The little measurement there is points towards using the tool to calibrate feel, not to replace it. In a 1998 Ukrainian study the archer estimated their own release time after each arrow and compared it with the reading from a device on the clicker, and by the end of the method the release timing was more regular and the competitive result had improved; in 2016 an experienced archer judged as poor some shots that the sensor recorded as good. The data are thin, and the 1998 study does not even give the number of archers, but the principle is the one I use with my own archers: feel first, then the comparison with the measurement.
The other risk is access. The reference tools, such as multi-camera optical systems and shooting machines, are found mostly in laboratories and federation centres, while those within everyone's reach, such as video apps and wristbands, have been tested on few archers and, for fine measurements such as angles, without an error that can be read. Archers training on limited means may end up trusting the least checked numbers, which is one more reason to ask any app how large its error is before using it to change your technique.
A personal closing thought
What draws me to archery, after years of shooting and studying it, is the meeting between an ancient movement and tools that only in the last few years have been able to measure it outside a laboratory. That is exactly why I prefer dated examples to forecasts: a 2024 measurement with its error says more than a promise for 2040. The registry entries linked from this article are updated as new studies arrive, and when one of those studies changes the status of an entry I update this page too.