Research On Mechanism Optimization Design And Grasp Planning Of Tendon-driven Multi-finger Dexterous Hand

With the development of the Chinese space exploration project,requirements of space missions in many fields are increasing continuously,such as on-orbit maintenance,space equipment assembling and planet exploration.However,because of the severe space environment,astronauts are hard to stay in the outer space for a long time and to discover unknown space areas.Therefore,utilizing robotic astronauts instead of human astronauts is an important research topic of various space missions.Meanwhile,as the end effector of the robotic astronaut,anthropomorphic dexterous hand is the key to realize multifunctional operations.Because of the complex structure of human hands,completely imitating human hand's structure leads to high complexity and instability of robotic hands.Optimizing the robotic hand system to decrease the complexity of the mechanism and to realize the light-weight structure has become an important topic in the design of space anthropomorphic hand.Moreover,in order to adapt to the complicated and changeable space environment,robotic hand is required to be able to independently execute manipulation tasks in unknown environment.Therefore,optimal mechanism design of anthropomorphic hand and planning methods to grasp unknown targets are studied in this paper.Firstly,the qualitative optimization of system topology structure is discussed based on analyses of human hand's anatomical structure and manipulation postures.A design scheme of a four-finger dexterous hand with low complexity and necessary dexterity is proposed.The carrying capacity of each finger,the motion property of each joint and human hand's degree of freedom configuration are discussed according to the skeletal strength,the distribution of muscle tendon and the skeletal properties of finger joints.Diverse classification methods of hand postures are introduced based on three classical hand posture theories.Properties and joint coupling modes of different operation postures are studied.Synthesizing above,the structure of human hand is divided into the major and the minor parts for manipulation and bearing.Meanwhile,the minor joints are coupled by principal component analysis of human posture's data.In order to obtain appropriate mechanism parameter values,a parameter optimization method based on the evaluation of dexterous hand's manipulation capabilities is proposed.The optimization process is divided into evaluation of single finger dexterity and multi-finger collaboration performances.The single finger optimization analyzes the velocity transfer characteristic from the joint space to the Cartesian space.The optimization objective function is formulated by minimizing the conditional number of single finger's Jacobian matrix.For multi-finger optimization,series of normalized evaluation indices are proposed to evaluate the grasp and manipulation workspaces based on the standard evaluation gestures extracted from three basic manipulation missions of dexterous hand.Final optimization result is acquired by kinematics analysis of the relations between the mechanism parameters and the evaluation indices.However,because the analytical estimation costs a large amount of computation and is susceptible to the variance of targets and joint angles,a fast graphing evaluation method based on Monte Carlo method is proposed.The feasibility of the graphing method is discussed by comparing with the analytical method.By synthesizing above optimization analysis,a novel light-weight four-finger anthropomorphic dexterous hand is proposed,which can grasp multidimensional targets and has large in-hand operation space.Besides,by comparing with the last generation of dexterous hand,various issues in dexterous hand development are discussed,such as complex three dimensional mechanical design,flexible tendons arrangement,as well as system maintenance and expansion.The first-generation dexterous hand utilizes five-finger structure,guideless tendon driven system and one-piece structural components.The secondgeneration dexterous hand has a four-finger structure configuration,whole-guidance tendon driven system,multi-piece structural components,as well as the optimized system structure and mechanism parameters.Meanwhile,by comparing the performance evaluation of two dexterous hands,the main influence factors of system manipulation capabilities in mechanism design are comprehensively analyzed.Finally,a precise grasp planning method to grasp unkown objects based on optimization theories are proposed,which can realize the high dimensional joint space and the base frame posture planning of the dexterous hand simultaneously.Firstly,a superquadric fitting method is utilized for reconstructing unknown object models based on object's point cloud.The fitting result for irregular objects is improved by the pre-processing of point cloud segmentation and the post-processing of fitting detection.Secondly,a “local-global” optimization method is proposed for precise grasp planning.This method searches valid grasp postues in open space and is not limited to certain approach direction or preset gesture dataset.The local planning optimizes the contact condition between finger tips and model surface.The global planning based on particle swarm optimization optimizes the distribution of contact points and searches for more stable grasp postures.Finaly,grasping tests are executed in the simulation environment,and the applicability and properties of this planning method are analyzed in detail.The results shows that the successful ratio of grasping is preferable for targets with global or local regular convex shapes.