Braille Recognition Algorithm Based on Robotic Arm Active Tactile Perception

With limited Braille literacy among the visually impaired, assistive technologies that do not depend on prior Braille knowledge are urgently needed. This study presents a Braille recognition framework using PyBullet simulation and a KUKA iiwa robotic arm to actively acquire tactile data from Braille characters. A deep neural network (DNN) is trained on 19-dimensional features combining statistical and temporal force signals. The dataset includes 5,000 custom Braille characters based on a 4×4 dot matrix and 2,000 Chinese Braille characters. The system captures both normal and tangential forces during contact. Experimental results show that recognition accuracy is up to 92 % across both datasets. The proposed method is low-cost, repeatable, and effective for tactile Braille recognition, offering practical potential for blind assistance and Braille education in resource-constrained environments.