Decoupling Research Of Three-dimensional Force Flexible Tactile Sensor Array Based On Neural Network Methods

With the rapid development of intelligent robot technology, research and development of flexible multi-dimensional force tactile sensor has become one research focus of robot skin. The flexible tactile sensor is a very important part to improve the intelligent level of robots. It is required that the flexible tactile senor should have softness and flexibility as human skin, be suitbale for surfaces of different roughness and shape, and quickly and accurately achieve the information acquisition from external environment. Especially, the intelligent tactile perception of robot is critical for ensuring the safe and efficient interaction between robot and outside world. Therefore, the research of skin-liked flexible multi-dimensional force tactile sensor plays an irreplaceable role in the bionic intelligent robot field.In order to solve the difficulties in the modeling process, decoupling procedure for the flexible tactile sensor and high nonlinearity of the sensor model, we carry out a series of research on the flexible tactile sensor modeling and multi-dimensional information decoupling, with the sensor technology, artificial neural network (ANN), elastic mechanics, numerical algorithm and other various theoretical approaches. This paper concentrate on the study of the multi-dimensional information decoupling problem and try to approximate the high dimensional nonlinear mapping relationship of the tactile sensor array using various ANNs, so as to improve the decoupling accuracy and real-time performance. The main research contents and innovation achievements are as follows:1. A decoupling method for the tactile sensor array based on Back Propagation Neural Network (BPNN) is proposed, and the three-dimensional (3-D) deformation is decoupled from the resistance information. Firstly, the BP algorithm is improved, and the BP networks are constructed with different hidden layer nodes respectively. Secondly, k-fold cross-validation (k-CV) method is applied to construct the dataset. Experimental results show that k-CV method can effectively improve the decoupling accuracy of the sensor. Finally, the deformation of tactile sensor array at different scales is decoupled by the BPNN, which significantly increases the decoupling accuracy and demonstrate the efficiency and feasibility of the method.2. A new decoupling method based on Radial Basis Function Neural Network (RBFNN) for the three-dimensional (3-D) force flexible tactile sensor is presented. A numerical model of the tactile sensor is built by ANSYS finite element analysis software, which simulates the mapping between different3-D force applied on6equal areas and deformation of the sensor. Furthermore, the RBFNN is improved by k-means and the least square method (LSM). Then it is applied to decouple the nonlinear relationship from the deformation to the3-D force. At last, the high dimensional nonlinear mapping relationship between the resistances and3-D force is decoupled by the improved RBFNN algorithm directly. The research results show that the improved RBFNN with high nonlinear approximation ability has good performance in decoupling3-D force and satisfies the decoupling accuracy requirements of the flexible multi-dimensional force tactile sensor.3. An efficient method is proposed to simulate the deformation of a flexible tactile sensor interfered by noises in the practical application. White Gaussian noises are added into the ideal tactile sensor model. Then the optimized RBFNN is used to approximate the nonlinear mapping relationship between the resistances with white Gaussian noises and3-D deformation. The3-D deformation is decoupled by the resistance information.4. Theoretical analysis and verification is performed on the structure and principle of a novel3-D flexible tactile sensor, and to deduce the mapping relationship between the resistances and3-D force. Then, the force applied on the tactile sensor is decoupled from the resistance information based on BPNN through the optimized number of hidden layer nodes. The new flexible tactile sensor achieves the decomposition of the three-dimensional information from the structure with its unique design, avoids the direct interference between the nodes, reduces the complexity of the sensor model and the degree of nonlinearity, and then accelerates the decoupling rate.The above research achieves the efficient decoupling for the3-D deformation and3-D force of the3-D force flexible tactile sensor, improves the decoupling accuracy, and provides the theoretical basis for further research on multi-dimensional flexible tactile sensor array.