High-Speed Tactile Braille Reading

Vision-based tactile sensing system for robotic Braille reading.

Overview

Designed and built a high-speed vision-based tactile sensing system for robotic Braille reading, inspired by human sliding touch. The system achieves fast and accurate character recognition by leveraging continuous motion rather than static probing.

High-speed tactile braille reading via biomimetic sliding interactions Potdar, P., Hardman, D., Almanzor, E., Iida, F.
IEEE Robotics and Automation Letters (RA-L), 2024

Key Contributions

  • Developed a biomimetic sliding interaction pipeline for tactile perception
  • Built a real-time inference system for sequential Braille character recognition:
    • Deblurring Autoencoder to remove motion blur from sliding touch images
    • YOLOv8 Object Detection to localise and classify Braille characters in the deblurred images
    • Data-driven Error Correction to improve accuracy by learning common misclassification patterns
  • Demonstrated robust performance under motion and sensing noise

Results

  • 315 words per minute reading speed
  • 87.5% character-level accuracy
  • Significant speed improvement over prior static tactile methods

Notes

This work highlights the importance of active perception and motion-driven sensing for efficient robotic interaction.