Researchers at the University of Cambridge have achieved a groundbreaking development in the field of robotics by creating a robotic sensor equipped with artificial intelligence (AI) techniques capable of reading Braille at a speed approximately twice that of most human readers. The research team, led by Professor Fumiya Iida, utilized machine learning algorithms to teach the robotic sensor to swiftly navigate lines of Braille text, achieving an impressive reading speed of 315 words per minute with close to 90% accuracy.
While not initially designed as assistive technology, the high sensitivity required for Braille reading positions the robotic sensor as an ideal testbed for the advancement of robot hands or prosthetics with comparable sensitivity to human fingertips. Human fingertips possess remarkable sensitivity, detecting subtle changes in texture and enabling precise object manipulation, a feat challenging to replicate in robotic hands efficiently.
Parth Potdar, the paper’s first author and an undergraduate at Pembroke College, emphasized the importance of softness in human fingertips for optimal gripping, a characteristic that proves challenging to integrate with high sensor information in robotic hands, especially when dealing with flexible surfaces. Braille, with its closely spaced dots, serves as a perfect test for the robotic ‘fingertip,’ demanding high sensitivity.
The team employed an off-the-shelf sensor to create a robotic Braille reader that emulates human reading behavior more accurately than existing robotic Braille readers. Unlike conventional readers that work in a static manner, touching and reading one letter pattern at a time, the new robotic sensor slides efficiently along rows of Braille characters, demonstrating a more realistic and efficient approach.
To address the challenge of motion blur in the images captured by the robotic sensor’s camera, the researchers developed machine learning algorithms capable of “deblurring” the images before letter recognition. Training the algorithm on sharp images with artificially applied blur, the team achieved remarkable results, with the robotic Braille reader reading at 315 words per minute with 87% accuracy. achieving a speed double that of a human Braille reader and approaching a level of accuracy comparable to theirs.
David Hardman, co-author and researcher in the Department of Engineering, highlighted the achievement’s significance, emphasizing the trade-off between speed and accuracy observed, mirroring the characteristics of human Braille readers. The research findings could have broader applications beyond Braille, potentially impacting areas like detecting surface textures or slippage in robotic manipulation.
Looking ahead, the researchers aim to scale the technology to the size of a humanoid hand or skin, with support from the Samsung Global Research Outreach Program.
