Under-actuated hand grasping classification method based on pneumatic tactile sensation

Pneumatic tactile array sensors were used to pick up tactile feature information. The palm and fingertips were equipped with tactile sensors, and their sensitivity and maximum repeatability error are 13.73, 10.85 kPa/N and 3.54% and 2.73%, respectively. Tactile feature data sets which were built by under-actuated used discrete Kalman filtering method to reduce noise, and were classified by random forest algorithm. Experiment results indicate that the pneumatic tactile sensor has high sensitivity and small repeatability error, and can effectively sense the changes of tactile information during the capture and classification process, relying on the under-actuated hand, avoiding the impact of speed collision on the classification results during grasping, and without complicated grasping control methods. The random forest classification method combined with discrete Kalman filtering can effectively complete the recognition and classification of objects. The recognition accuracy of this method is 93.2%. © 2021, Editorial Board of Journal of Huazhong University of Science and Technology. All right reserved.