⚡ electrical engineering

Speed-Accuracy Trade-Off in VR-based Eye-Hand Coordination Training: The Impact of Task Instructions and Physical Abilities

This paper challenges the assumption that throughput is an invariant metric for evaluating VR-based eye-hand coordination training by demonstrating through a longitudinal study of 36 participants that task instructions and individual strategies significantly influence the speed-accuracy trade-off, thereby revealing throughput's limitations as a reliable indicator of learning progress.

Original authors: Nour Hatira, Anil Ufuk Batmaz, Mine Sarac

Published 2026-09-29

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Nour Hatira, Anil Ufuk Batmaz, Mine Sarac

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a world where training your body to move with precision happens not on a sports field or in a gym, but inside a computer-generated space. This is the realm of virtual reality, a technology that has moved beyond entertainment to become a serious tool for teaching the brain and body how to work together. At the heart of this training is a concept known as the speed-accuracy trade-off. It is a simple but powerful idea: when you try to do something very fast, you are more likely to make a mistake; when you try to be perfectly precise, you usually have to slow down. For decades, scientists have used a specific mathematical rule to measure how well people perform these tasks. This rule assumes that no matter how a person balances speed and accuracy, their overall performance score should stay the same, like a fixed line on a graph. Researchers have relied on this assumption to judge whether people are getting better at tasks like catching a ball or operating a robotic arm. But what happens when people train for days, not just minutes? Does that old rule still hold true when the brain is actively learning and adapting?

A team of researchers at Concordia University and Kadir Has University set out to test this very question. They wanted to see if the standard way of measuring performance in virtual reality training was actually telling the whole story. To do this, they brought together 36 people and asked them to play a game for eight consecutive days. The game was simple: wearing a virtual reality headset, participants had to reach out with their hands and touch glowing spheres that appeared on a grid. The goal was to touch them as quickly and accurately as possible. The researchers were interested in how different types of people learned this task. They divided the participants into three groups: healthy individuals with no special training, professional volleyball players who are used to reacting quickly, and a third group of healthy people who wore a one-kilogram weight on their wrist to simulate the movement limitations of someone with a physical impairment.

To see how instructions changed the way people learned, the researchers gave each group of participants a different set of rules. Some were told to focus entirely on being accurate, even if it meant moving slowly. Others were told to focus on avoiding mistakes, while a third group was given no specific instructions at all and told to play however they liked. Over the course of eight days, the researchers tracked every movement, recording how long it took to touch a target and how close the touch was to the center. They used the standard method of calculating a single score that combined speed and accuracy, known as throughput, to see if it could capture the progress of the learners.

The results revealed a surprising flaw in the old way of thinking. When the researchers looked at the data day by day, they found that the participants did not follow the expected pattern. The standard rule suggested that if someone got faster, they should get less accurate, and if they got more accurate, they should get slower. The researchers expected the overall score to stay steady as people shifted between these two extremes. Instead, they found that most people got better at both speed and accuracy at the same time. They did not have to sacrifice one for the other. As the days passed, the professional athletes became faster, and the people with the weighted wrists became more precise, yet both groups improved their overall performance without hitting the trade-off wall that the old rule predicted.

The study also showed that the type of instruction mattered deeply, but not in the way the single score suggested. For the healthy participants, telling them to focus on accuracy helped them become more precise, but telling them to focus on speed made them faster, often at the cost of precision. However, for the professional athletes, the instructions had a different effect. Even when told to be careful, they remained fast, likely because their years of sports training had made quick reactions a habit. The people with the weighted wrists moved slowly, but they were incredibly consistent, showing that physical constraints can force a person to adopt a different, highly stable strategy.

Perhaps the most important finding was that the single score used to measure performance was actually hiding these differences. Because the score combined speed and accuracy into one number, it looked like everyone was improving in the same way. It failed to show that one group was getting faster while another was getting more precise, or that some people were improving in both areas simultaneously. The researchers concluded that this single score is not a reliable way to measure learning over time. It assumes that speed and accuracy are locked in a fixed balance, but the human brain is more flexible than that. When people train for days, they can often improve both dimensions together, breaking the old rules.

This study suggests that to truly understand how people learn new motor skills in virtual reality, we need to look at speed and accuracy separately. We cannot rely on a single number to tell us if someone is getting better. Instead, we must watch how their speed changes and how their precision changes, and see how those two things interact. For designers of training programs, this means that giving people specific instructions can shape exactly how they learn. If you want a trainee to be more precise, you must tell them to be precise, and they will likely slow down to achieve it. If you want them to be fast, they will speed up, but they might make more mistakes. The old assumption that performance is a fixed line is wrong; the path to skill is a journey where speed and accuracy can grow together, and the way we measure that journey needs to change to reflect the reality of human learning.

Technical Summary: Speed-Accuracy Trade-Off in VR-based Eye-Hand Coordination Training

Problem Statement
Virtual Reality (VR) has emerged as a viable platform for Eye-Hand Coordination Training (EHCT), offering immersive environments for motor skill development. While previous research has established Fitts' Law and effective throughput as standard metrics for assessing performance in these tasks, a critical assumption underpins their use: that throughput is invariant to the speed-accuracy trade-off. This assumption, largely derived from single-session experiments with controlled cognitive sets, suggests that throughput remains constant regardless of whether a user prioritizes speed or accuracy. However, the literature lacks empirical investigation into how these metrics behave during longitudinal learning curves, particularly when participants have varying physical abilities and are subjected to different task instructions. It remains unclear whether throughput can accurately reflect motor learning progress when users adapt their strategies over extended training periods, or if the metric obscures distinct learning patterns driven by the speed-accuracy trade-off.

Methodology
To address these gaps, the authors conducted a longitudinal user study involving 36 participants over an 8-day period (1-day pre-training, 6-day training, 1-day post-training). The study utilized a custom VR EHCT system developed in Unity, featuring a 5×6 grid of spherical targets. Participants interacted with the system using an Oculus Quest 2 HMD, selecting randomly activated targets as quickly and accurately as possible.

The study employed a 3x3 factorial design:

Participant Groups (Physical Abilities):
- Healthy: Individuals with no specialized physical training.
- Athletes: Professional volleyball players with advanced eye-hand coordination skills.
- Simulated Limited-Movement: Healthy participants wearing 1-kg weights on their dominant wrists to simulate movement constraints and physical impairments.
Task Instructions (Motivations):
- Accuracy-Focus: Instructed to prioritize hitting the center of targets, even if it meant slower speeds.
- Error-Focus: Instructed to minimize selection errors (hitting the wrong target).
- No-Focus: No specific instructions; participants performed the task freely.

Evaluation Metrics
The study analyzed standard Fitts' Law-derived metrics, including Execution Time, Error Rate, Accuracy (standard deviation of touch points, $SD_x$ ), Precision (effective index of difficulty, $ID_e$ ), and Effective Throughput. Throughput was calculated as the ratio of $ID_e$ to mean execution time.

Key Results

Divergent Learning Patterns: The interaction between physical ability and task instruction significantly influenced performance.
- Simulated Limited-Movement: Consistently achieved the highest accuracy and precision but the slowest execution times. Under accuracy-focused instructions, they significantly improved accuracy without changing execution time. Under other conditions, they showed co-improvement in both speed and accuracy.
- Athletes: Consistently demonstrated the fastest execution times. Under accuracy-focused instructions, they improved speed significantly but showed no significant gain in accuracy, suggesting their established motor strategies were resistant to instruction-based changes.
- Healthy: Showed the most variability based on instruction. Under accuracy-focus, they improved accuracy without speed gains. Under no-focus instructions, they exhibited a clear speed-accuracy trade-off, becoming faster at the cost of accuracy.
Throughput Invariance Failure: Contrary to the assumption that throughput is invariant to the speed-accuracy trade-off, the study found that throughput varied significantly based on task instructions and individual strategies.
- Throughput increased significantly for all groups over the 8 days, regardless of whether the underlying learning trajectory involved co-improvement (speed and accuracy both increasing) or a trade-off (one increasing while the other decreased).
- The metric failed to distinguish between qualitatively different learning patterns. For instance, it masked the fact that athletes improved speed but not accuracy under specific instructions, while the limited-movement group improved accuracy but not speed.
Violation of Constant-Throughput Assumption: By constructing theoretical constant-throughput curves based on Day 0 baselines and overlaying them with longitudinal data, the authors demonstrated that participant performance did not follow these curves. Instead of the compensatory trade-off (slowing down to gain accuracy) predicted by Fitts' Law, most groups improved both speed and accuracy simultaneously. This indicates that the speed-accuracy trade-off weakens or disappears during longitudinal motor learning.

Key Contributions

Longitudinal Analysis of EHCT: The study provides empirical evidence on how speed-accuracy trade-offs evolve over multiple days of training across diverse physical profiles, a gap in existing literature which has focused primarily on single-session tasks.
Critique of Throughput as a Learning Metric: The paper challenges the validity of using effective throughput as a holistic indicator of learning progress in VR-based EHCT. It demonstrates that throughput is not invariant and can obscure distinct learning strategies and physical constraints.
Impact of Task Instructions: The research highlights that task instructions (accuracy vs. error vs. no-focus) interact with physical abilities to shape motor learning trajectories, suggesting that a "one-size-fits-all" metric is insufficient for evaluating training efficacy.

Significance and Claims
The authors claim that their findings reveal inherent limitations in using throughput as a reliable indicator of learning progress, particularly in scenarios characterized by strong speed-accuracy trade-offs or longitudinal adaptation. The study concludes that Fitts' Law and its derived throughput metric, originally designed for single-session controlled tasks, may not fully reflect the dynamics of motor learning over time. Consequently, the paper argues that evaluating EHCT progress requires separate and combined analyses of speed and accuracy rather than relying on a single composite metric. The work suggests that alternative performance metrics are needed to more accurately capture user adaptation and learning in VR-based training environments.

Technical Summary: Speed-Accuracy Trade-Off in VR-based Eye-Hand Coordination Training

More like this