A Research On The Mechanism Of Vision Based Tactile Sensing For Adaptive Flexible Grasping

The development of robots provides great convenience for people's daily life and industrial production.The gripper as a commonly used end effector for robots has always been one of the research focuses in the field of robots.In recent years,the development of soft robot promoted the research of soft grippers.Compared with traditional rigid manipulators,soft grippers relying on their flexibility of materials and structures will cope with unstructured environments better and improve interactive security.As one of them,the self-adaptive structured finger can realize the covering of the gripping target autonomously due to its own structure's "intelligence" and reduce the gripping control requirements,which has been widely used.However,the existing research on the adaptive structure finger still has a big gap in the study of the adaptive finger's own mechanical characteristics and the realization of the external tactile sensing.Therefore,in this thesis,the mechanical characteristics of the adaptive structure finger were measured in detail through the experimental method,and the concept of tactile image sensing was introduced into the field of soft tactile sensing.The advancement of the finger affects the accuracy of the prediction of the contact position and normal contact force with vision-based tactile sensing method.In this thesis,we analyzes the state of the art of soft adaptive grip,exteroception of soft gripper,and vision-based tactile sensor.We finds that there are gaps in the research on adaptive fingers.Through a comparative analysis of the existing soft tactile sensing and vision-based tactile sensor,the advantages of the vision-based tactile sensor for adaptive gripping applications are summarized.Besides,the concept of adaptive tactile fingers based on vision is proposed.Firstly,based on simulation experiments,the distribution of the payload and equivalent stiffness of a typical adaptive finger structure is measured by surface and curve fitting,and the distribution is analyzed in detail through qualitative and quantitative methods to discuss the effect of contact conditions.Then,the omni-directional adaptive finger with configuration of 3D grid grid is proposed,and the test platform is designed to measure the contact force in different load state.In the physical environment,with specific finger structure as the object,the payload and the distribution of equivalent stiffness were measured in detail to verify andsupplement the simulation experiment results.Lastly,a complete framework of learning-based visual-tactile fusion system is proposed.With the omni-directional adaptive finger structure as the object,the design of the tactile finger was completed,and its performance was tested experimentally.The results show that the finger can simultaneously predict the contact position and 2-D contact force information during grasping.The accuracy of the position prediction is up to 90%,and the root mean square error of the contact force prediction is 0.820 and 0.355,respectively.