Counterfactual Badminton Motion Guidance

IEEE ICRA Workshop on Wearable Robotics · 2024 · Lead author

What could a different version of this movement look like?

Copying an expert’s motion does not directly show how a player’s own movement might change. This workshop study explores nearby alternatives that can be compared with the original stroke.

Published pipeline: joint positions, autoencoder latent search, stroke-quality classifier, and an overlay of original and generated motions — The existing system overview connects model-based motion generation with visual comparison.

System

MultiSenseBadminton joint-position sequences feed a stroke-quality classifier. An autoencoder-based search generates counterfactual motions in latent space. A Unity visualization overlays generated and original movements.

Evidence and limits

The 2024 paper compares 100 examples per stroke type with three nearest-neighbor baselines. It evaluates model-based proximity and motion distances. These offline results do not establish that the suggested motions improve a person’s performance or learning; user evaluation is future work.

Resources

Paper title: Counterfactual Explanation-Based Badminton Motion Guidance Generation Using Wearable Sensors.

Back to publications