The Design Of Anthropomorphic Robotic Hand And Research On Hand Pose Estimation

Early robots already had the ability to move heavy loads quickly and accurately,and to program flexibly.However,the terminal mechanism needs to be designed according to the specific application.Due to the lack of flexibility of the terminal mechanism,there are many limitations in the application of different types of tasks.With the improvement of computer computing power and the miniaturization of electronic devices,it is possible to develop a universal end-effector with hand shape.These devices can capture and manipulate objects like humans.The design of this manipulator is relatively complex,because dexterity means that a large number of degrees of freedom are required in limited space,and a large number of suitable drivers and sensors need to be integrated on the manipulator.Some improvements are also needed in the control of the manipulator to make the control easier.It's intuitive for us to use gestures to communicate.One way to realize gesture recognition is to track the position of the hand and then detect the gesture of the hand.Detecting and tracking the bare hand's centroid position is simpler than the exact configuration of detecting joint position.However,most applications require accurate pose detection and stable tracking.Even with today's powerful processors,the extraction of hand joint posture remains challenging.In this paper,a anthropomorphic robotic hand is designed and its motion is controlled by gesture recognition method.The anthropomorphic robotic hand designed in this paper has a similar shape and size to the human hand.The main driving force is provided by 12 DC motors,and the main driving force is used to complete the flexion movement of the finger.The stretching action of the finger is driven by the passive force provided by the elastic rope.The driving motor is placed on the palm and back of the hand,and the power is transmitted by flexible rope.In order to make the object of the anthropomorphic robotic hand,this paper uses the popular 3D printing technology to make the parts of the anthropomorphic robotic hand.Because it has more controllable degrees of freedom,the anthropomorphic robotic hand has high flexibility and can imitate the action of human hands.In the bottom control of the anthropomorphic robotic hand,the controller of the anthropomorphic robotic hand is a 32-unit microcontroller.The PID algorithm is used as the bottom control algorithm of the motor,and the rotary potentiometer is used as the joint angle sensor.A servo system for driving anthropomorphic robotic hand is composed of a controller based on PID algorithm,a decelerating DC motor and a joint angle sensor.In order to facilitate the control,the hand pose recognition system composed of Kinect depth camera and depth neural network is used to estimate the hand pose in real time and control the motion of the anthropomorphic robotic hand with the estimated results.The depth camera can get the depth information of the scene.After the hand is segmented from the depth image,the estimated gesture of the hand can be obtained from the output of the depth neural network by inputting the segmented depth image into the trained depth neural network.It is assumed that the hand is the closest object to the depth camera when separating the hand.The depth neural network is designed according to the task of estimating the gesture of the hand.A priori knowledge is added to the design and training,and an optimization program is added to the estimation process.Before controlling the actual anthropomorphic robotic hand,the virtual debugging is carried out on the platform of Robot Operating System(ROS).Using the Robot Visualization Module(rviz)in ROS,the model of the anthropomorphic robotic hand and the effect of real-time driving the motion of the anthropomorphic robotic hand can be displayed on the display.Virtual debugging greatly reduces the possible damage in debugging,thus improving the efficiency of debugging.