| Haptic perception is a prerequisite for robots to perform fine manipulation tasks that help them interact with and reflect the physical properties of objects.Haptic sensors can provide robots with richer information that enables them to perform a wider range of tasks.Therefore,it is important to analyze and study the acquisition,processing and utilization of haptic information to enhance the perceptual capabilities of robots and improve their intelligent interaction.In this paper,based on the topic of underactuated manipulators grasping object recognition by machine learning,a robotic grasping object recognition experimental platform was formed,which contained an underactuated manipulators hand and a flexible capacitive sensor data acquisition system.The flexible capacitive sensor,the acquisition of capacitive information,and the training and testing of machine learning models were investigated.For the problem of data set acquisition,a planar capacitive sensor with microstructure was designed and prepared,and the rationality of the design was verified by COMSOL simulation.The data were processed using Kalman filtering to reduce the influence of the environment on the data.For the problem of recognition of grasping objects based on capacitive data,flexible capacitive sensors were used to obtain data sets.These datasets were used to train and test the machine learning model.The dataset was expanded and features were extracted using fast Fourier transforms to increase the size and diversity of the dataset.An experimental platform was built and grasping experiments were conducted for the underactuated manipulators grasping recognition problem.In the experiments,four different objects were selected for grasping,and a machine learning model was used to recognize the shape of the grasped objects.The recognition results of the model were further analyzed,and some problems of the model were found through error analysis,and corresponding solutions were proposed. |
PDF Full Text Request