Bao Guanjun

Pneumatic Robot Multi-fingered Dexterous Hand - ZJUT Hand

For the shortcomings of existing dexterous hands,such as structure complexity and control difficulty,a new type of pneumatic multi-fingered dexterous hand,named ZJUT Hand,is proposed.Based on a flexible pneumatic actuator (FPA),pneumatic stiffness/compliance bending joint and side-sway joint are designed.After that,a 4-DOF pneumatic anthropomorphic finger with four joints is assembled.To achieve a high degree of modularity,five identical fingers are installed on a anthropomorphic palm to constitute the body structure of ZJUT Hand with 20 DOFs.The structural parameters of ZJUT Hand are determined by the bionic optimization method,and grasping simulation experiments are carried out for body structure of ZJUT Hand.Simulation results show that:ZJUT Hand can grasp objects with typical shapes such as cylinder,strip and spherical objects,and achieve the pinching,gripping,pulling and other complex movements like human hand.The force/position sensory system for ZJUT Hand is developed.The grasp planning experiments for ZJUT Hand is completed.Experimental results show that ZJUT Hand is able to perform stable grasping for typical target objects.Finally, the characteristics of ZJUT Hand are summarized.

Cucumber image segmentation algorithm based on rough set theory

Abstract In this paper a segmentation algorithm for cucumber images is presented to solve the problems of target identification and position for a greenhouse cucumber‐harvesting robot. Amethod of combining the three basic colours, branch‐quantity and threshold processing is used to enhance the target im ages, increasing the contrast between background and target and reducing the information content of the colour image. The strengthened image is considered as a knowledge system. The cucumber images are then segmented by a classification decision table which is based on a knowledge‐reduction algorithm from rough set theory, choosing the image colour attribute as a condition attribute and standard colour as a decision attribute. The image noise is removed using an expansion algorithm of mathematical morphology. The experimental results show that the proposed algorithm can enhance the image of greenhouse cucumbers and segment the profile of the cucumber image effectively. There is less noise in the image after segmentation.

Grasping model identification for ZJUT hand

A new type of air-driven multi-fingered dexterous hand, named ZJUT Hand, is proposed. ZJUT Hand has good passive flexibility and can make up for the deficiency of existing dexterous hands in some ways. A new method for grasping model identification based ANFIS is proposed. Simulation experimental results show that the method can easily establish the model of the target objects, and has very good recognition accuracy and convergence. Equivalent rules of target objects are presented and greatly improve the efficiency of identification. A general grasp planning program of multi-fingered dexterous hand is proposed. Finally, the grasp planning experiments for ZJUT Hand is completed. Experimental results show that ZJUT Hand is able to construct the models of two typical target objects.