Research On Design And Grasping Planning Of Anthropomorphic Five-fingered Dexterous Hand

The anthropomorphous robotic dexterous hand, as the end effectors of the humanoid robot, is an important part which improves the executing capability and intelligent levels. It is always a hot research that how to design the real anthropomorphous dexterous hand in both shape size and manipulation capability, and make the designed dexterous hand have the ability of grasping arbitrary object. Based on national high technology research and development program"Research on structure design and grasping manipulation of new five-fingered dexterous hand", the dexterous hand which is the international advanced level in shape and manipulation capability is developed.The design pivot is focused on how to design the dexterous hand with enough power, higher integration and interchangeability, close to human hand size and having commercialization value based on current commercial devices in this dissertation. HIT/DLR Hand II dexterous hand with 15 Degree Of Freedoms (DOFs) is composed of an independent palm and five identical modular fingers, in which each finger is composed of the based joint unit with 2 joints and the finger unit with 2 joints and 1 DOF. All the systems including the motor, the transmission agent, the sensor and electric in each unit are respectively integrated in the finger body. This design idea which each unit is modularity not only reduces the designing period, but also increases the integrated level. In addition, in order to make the dexterous hand have a better envelope space and the optimal grasping performance while the dexterous hand executes the grasping manipulation, the cambered palm and the thumb positions in the palm based on the human palm characteristic are designed by optimizing simulation analysis. The new-style drive mode which combines the micro-actuator, synchron timing belt and harmonic driver and the steel wire couple mechanism of finger end joint guarantee the shape size of the dexterous hand is similar to the human hand. The integrated design idea of HIT/DLR Hand II is embodied in not only the sensor structural body design but also the external packaging of the dexterous hand. The assorted sensors, including force/torque, position, temperature, tactile sense and so on, ensure the dexterous hand has higher survey accuracy without increasing the shape size. In the same way, the design idea integrated the package parts into framework parts.The super-powerful manipulation capability of human hand roots in not only the rather large manipulation space of 5 fingers but the ability of the human palm and the opposite thumb to adjust its configuration adaptively following different shape objects. Based on this, utilizing the definitions of three arches in anatomical field, the anthropomorphous arched palm is optimized designed. Integrating the manipulation space of the dexterous hand finger, the thumb position in the palm is optimized using interaction performance index between the thumb and the other fingers. According the above structure characteristic of the palm configuration, the grasp space of the dexterous hand is described mathematically and the kinematics model of the two finger close chain grasp is constructed. The mapping relationship between the fingertip and the object barycenter is described based on geometric method and the kinematics equations of each fingertip position in the global coordinate system are proposed.The uncertainty of manipulation space is one of the significant characteristics for the multi-finger dexterous hand grasping. For an unknown grasping task, the visual or laser scan technology are firstly utilized to make it known, and then the mathematics model of the grasped objects is estimated using corresponding algorithms in order to realize stable grasp. Taking the point contact with friction as the research object, the optimal grasp plane is firstly defined in this paper, and then the mathematics model of the grasped objects is established utilizing the super-ellipses parametric equation combining with the dexterous hand grasping tasks based on this plane. Grounded on this mathematics model, a model based on the Expert system and the Adaptive Neural Fuzzy Inference System (E-ANFIS) is proposed to reconstruct the grasped object model. This model does not need any visual aids under the situation with the foregone position and orientation of the manipulation objects, and could reconstruct the mathematics model of the grasped objects automatically only using the position relationship among the contact points with the five fingers. The classification between linearity and non-linearity of the manipulation object boundary utilizing the expert rules could increase the model precise and save online estimation time. The model presents the favorable convergence for the training data collected from the HIT robot arm/hand grasping system combining with virtual technology. Both the simulations and real grasping tests prove favorable anastomosis between the simulation and the real test results, and present a certain generalization capability.When the grasping model is known precisely, the next step is how to determine the grasping contact points satisfied with the Force Closure Grasp (FCG). A four-finger FCG algorithm based on the plane grasp is proposed aimed at the gait manipulation demands of HIT/DLR Hand II dexterous hand. The most significant characteristic of this algorithm is one of four contact points with FCG can lose the contact with the object without making the other three contact points losing the force-closure property.In the end, a robot arm/hand grasp manipulation experiment system and the virtual simulation manipulation environment are constructed. The training data collected from the simulation environment are used to train the fuzzy rule of the E-ANFIS model. In order to prove the machanical performance and the control precision of the dexterous hand body machanism, the joint position tracking and fingertip position tracking in Cartesian space are realized respectively based on the anti-integral-saturation PID control algorithm. The compliance experiments in free space and constrained space are operated for the fingers utilizing resistance control algorithms based on position. The shape recognition capability of E-ANFIS and the FCG algorithm rationality are verified with the grasping manipulation experiments utilizing the arm/hand system based on the above system.