Researchers at the University of Bristol presented an experimental environment and set of tasks designed to train AI to type on a Braille keyboard.
The Braille keyboard is a set of tactile symbols that blind people can use to enter information. Braille uses unique patterns of raised dots to represent alphabet letters or punctuation marks.
The general idea behind our work is to teach a robot to perform a task that humans perform manually. We also wanted to showcase the power of deep learning.
Nathan F. Lepora, one of the study authors
The authors of the work have developed four tasks, which involve typing on the Braille keyboard. To do this, the Researchers have created a simulated and real-world environment in which AI can learn to type in Braille. The researchers then trained state-of-the-art deep learning algorithms to perform the four tasks they created, in both simulated and real-world environments (that is, using a physical robot).
These deep learning algorithms have achieved excellent results: AIs have learned to perform all four tasks in simulations. Also, the result was successful three times out of four when implemented on a real robot. Only one of the tasks, which required the AI to continuously enter letters of the alphabet, proved difficult to translate.
The authors of this work were the first to successfully train AI to type on the Braille keyboard.