Abstract:The development of prosthetic hand technology has brought good news to the patients with hand loss, but most prosthetic hands lack tactile perception, while normal human hands are very sensitive to the physical characteristics of the contact object, so the human–computer interaction is insufficient. Aiming at the problem that the membrane pressure sensor commonly used in prosthetic hand tactile measurement is difficult to obtain and distinguish a variety of tactile information, an algorithm model framework based on deep neural network (DNN)-long-term and short-term memory network (DNN-LSTM) neural network is proposed. On the basis of recognizing the grip strength of the prosthetic hand, the system model adopts the inclined shaking tactile exploration method to identify the roughness of the grasping object. First, when the prosthetic hand grabs the object normally, according to the value of the pressure sensor, the grip strength can be identified through the DNN neural network, which avoids the defect of the measurement accuracy of the traditional pressure sensor. Then, through the tilting and shaking action, according to the change of the value of the pressure sensor, the roughness of the grasping object can be recognized through the D-LSTM fusion network. Finally, through model training and real object grasping experiments, the results show that the predicted results of the trained neural network model in real grasping are in good agreement with the preset results. In summary, our proposed tactile exploration action enables a single tactile perception source to obtain other tactile information, and the proposed D-LSTM model can effectively improve the recognition accuracy.

Robot Grasping Stability Prediction Network based on Feature-fusion and Feature-reconstruction of Tactile Information

Grasp Stability Prediction Using Tactile Information.

Grasp Stability Assessment Through Attention-Guided Cross-Modality Fusion and Transfer Learning

A Cascaded Deep Learning Framework for Real-time and Robust Grasp Planning

Predict Robot Grasp Outcomes Based On Multi-Modal Information

Visual-Tactile Fusion for Robotic Stable Grasping

Detecting Transitions from Stability to Instability in Robotic Grasping Based on Tactile Perception

Robotic Grasping Technology Integrating Large Kernel Convolution and Residual Connections

Integrating High-Resolution Tactile Sensing into Grasp Stability Prediction

Grasp Stability Assessment Through the Fusion of Visual and Tactile Signals Using Deep Bilinear Network.

Non-Matrix Tactile Sensors: How Can Be Exploited Their Local Connectivity For Predicting Grasp Stability?

Efficient Fully Convolutional Network and Optimization Approach for Robotic Grasping Detection Based on RGB-D Images

Grasp State Assessment of Deformable Objects Using Visual-Tactile Fusion Perception

A robot grasping detection network based on flexible selection of multi-modal feature fusion structure

Robotic Grasp Detection Network Based on Improved Deformable Convolution and Spatial Feature Center Mechanism

Flexible and robust robotic arm design and skill learning by using recurrent neural networks

Deep Learning Method for Grasping Novel Objects Using Dexterous Hands

CMG-Net: An End-to-End Contact-Based Multi-Finger Dexterous Grasping Network

Learning to Predict Friction and Classify Contact States by Tactile Sensor

Tactile Perception Information Recognition of Prosthetic Hand Based on DNN-LSTM

Visuo-Tactile-Based Slip Detection Using A Multi-Scale Temporal Convolution Network