EMG Fighting Game Gesture Control

ISEF Category: Technology Enhances the Arts

Subcategory: Games  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A forearm muscle signal can become a game input. That means one hand can trigger combos, not just buttons. For a player with limited hand use, that can change who gets to play at all. Your project tests whether the tech feels usable, fast, and fair.

What Is It?

Electromyography, or EMG, measures the tiny electrical signals your muscles make when they contract. A MyoWare sensor picks up those signals from your forearm, then software classifies them into gestures. Think of it like listening to different drum beats and guessing which one you heard. Each gesture can map to a move in a fighting game, like a punch, block, or combo starter.

A 1D-CNN, or one-dimensional convolutional neural network, is a machine learning model that works well on signal data like EMG. It looks for patterns across time, then assigns each signal to a gesture class. Your game turns those predictions into controls. The research question is not just whether the model works, but how well real players learn it and whether it helps one-handed players more than a standard controller setup.

Why This Is a Good Topic

This topic gives you a clear mix of coding, signal processing, and human factors. You can test something concrete, like gesture accuracy, reaction time, or how fast players learn the controls. It also connects to real accessibility problems in games, since many players cannot use two hands on a standard gamepad. A strong project here teaches you how to design a measurement system, compare user groups, and analyze learning over time.

Research Questions

• How does gesture recognition accuracy change as the 1D-CNN sees more training data? ?
• What is the effect of gesture type on false positive rate in the EMG game input system? ?
• Does one-handed players’ combo success rate improve faster than their gamepad baseline after repeated practice? ?
• To what extent does muscle fatigue reduce gesture classification accuracy during longer play sessions? ?
• Which gesture set produces the lowest confusion between similar forearm movements? ?
• What is the effect of sensor placement on gesture decoding accuracy and response time? ?

Basic Materials

• MyoWare EMG sensor or similar forearm EMG module.
• Arduino or similar microcontroller board.
• Laptop with Python installed.
• USB cable for data transfer and power.
• Fighting game prototype built in a simple game engine.
• Forearm strap or elastic band to hold the sensor in place.
• Standard gamepad for baseline testing.
• Timer or stopwatch for timing trials.
• Notebook or spreadsheet for recording trial results.

Advanced Materials

• Multi-channel EMG acquisition system.
• Shielded electrodes and lead wires.
• High-quality microcontroller or data acquisition board.
• Laptop with Python and a machine learning library.
• Game engine such as Unity or Godot for custom input mapping.
• External button box or instrumented controller for baseline comparison.
• Video camera for synchronized performance review.
• Force sensor or motion tracking device for supplemental validation.
• Statistical analysis software for repeated-measures testing.

Software & Tools

• Python: Cleans EMG data, trains the gesture classifier, and runs analysis.
• NumPy: Handles signal arrays and basic numerical processing.
• pandas: Organizes trial data and learning curve results.
• scikit-learn: Tests baseline classifiers and compares them to your CNN.
• ImageJ: Helps inspect video frames if you sync gameplay with motion clips.

Experiment Steps

1. Define the six gestures you will test and match each one to a game action.
2. Plan a data collection scheme that separates training, validation, and test samples.
3. Build a baseline model first, then compare it with the 1D-CNN.
4. Design a player study that measures accuracy, speed, and learning across repeated sessions.
5. Choose controls that isolate sensor placement, fatigue, and player experience.
6. Decide which statistics will compare EMG control against the gamepad baseline.

Common Pitfalls

• Training and testing on the same gesture samples, which makes the classifier look better than it really is.
• Letting the sensor shift on the forearm, which changes the signal and breaks consistency.
• Using gestures that look too similar, which drives up confusion between classes.
• Comparing EMG control to gamepad play without matching task difficulty, which makes the baseline unfair.
• Ignoring fatigue effects, which can hide a drop in accuracy during longer sessions.

What Makes This Competitive

A strong version of this project goes past a simple demo. You would compare multiple classifiers, test one-handed accessibility with real users, and report learning curves instead of a single accuracy score. You could also study whether gesture sets that feel easier to learn actually hurt performance later. Careful controls, repeated trials, and clear statistics would make the project much stronger.

Project Variations

• Test the same gesture-control system with a rhythm game instead of a fighting game to compare reaction timing.
• Compare forearm EMG input against webcam hand-tracking for one-handed accessibility.
• Study how well the classifier works when players use the system with fatigue, after repeated matches.

Learn More

• NIH PubMed: Search for review articles on EMG gesture recognition, human-computer interaction, and assistive gaming.
• NASA Open Data Portal: Find examples of signal processing and classification workflows you can adapt to your own analysis.
• MIT OpenCourseWare: Search for introductory machine learning and signals courses with free lecture notes.
• IEEE Xplore: Search abstracts on EMG-based game control, then use your school library for full-text access if available.
• arXiv: Search for recent preprints on 1D-CNNs for biosignal classification and compare methods.