Robot Grasp Operation Control Research Based On Tactile Information Perception

Grasping is one of the most frequently performed actions in human daily life.The intelligent development of robots can't be separated from the research of grasping operations control.The world is not a predictable assembly line,robots need plenty of perceptual information to support the work in unstructured environments.With the advance development of computer technology,some progress has been made in the application of visual and auditory perception of robots.However,the perceptual signals of robots would be disturbed and affected in complex unstructured environments.So that contact operations job are prone to failure,interaction security and other issues are caused.Tactile perception,as the only human contact perception ability,provides the necessary information for human beings in the process of contact the world.The same tactile perception system provides the possibility for robots to achieve grasping operations under the complex structural conditions when interact with the environment and targets.This paper focus on the relevant research of robot gripping operation control method based on tactile perception information.Primarily,according to the characteristics of human prejudice before gripping,the information of the working target was acquired in advance,and the convolution network was built to recognize the target information acquired by the image acquisition sensor.A robot platform based on manipulator and end-effector is built to adjust the desired output mode of gripping manipulation motion control by target identify information.For the existing gripping stability analysis methods,according to the advantages and disadvantages of the methods and the requirements of the robot gripping operation,the perceptual information processing method that meets the requirements was selected to judge the operation status of the robot gripping operation.The robot sensor system was constructed by choosing suitable working principle,the sensors was chosen to acquire desired tactile sensing information by signal handling method.The tactile sensing information was used to represent the state information of the robot in carrying out the clamping operation,and the perceptual output under different state information was analyzed.The sensor signal characteristics were compared in the diverse contact state,preparing the perceptual data for the robot to perform the clamping operation.In order to deal with the non-linear and complex tactile input information,the machine learning method was selected.The I/O model was trained by using radial basis function neural network for the desired output.The network structure parameters were optimized by improved subtraction clustering.Finally,the gripping operation experiments under the control of sensory information were carried out.The expected output of the robot's gripping operation was corrected by the pre-information of target recognition to optimize the established control framework.Experiments were carried out according to task objectives with different characteristics.The experimental acquisition results showed that the control method used could meet the task of gripping operation.