Mimic Human Grasping Control Of Multi-degree Of Freedom Robotic Arm

In 2021,the number of industrial robots installed worldwide has reached 517000,an increase of 31%over the previous year.As an important component of robots,robotic arms can assist or replace humans in completing various operational tasks and play an important role in industries,military,medical,aerospace,and other fields.Despite the widespread application of robotic arms,it is currently difficult to have both the high degree of freedom and dexterity of robotic hand and mechanical arm.The development environment of robotic arms and robotic arms is often independent of each other,lacking effective integration,and most of them still rely on tedious program commands for control.Compared with the natural grasping behavior of human hands,robotic arms have poor program portability and low generalization ability when facing complex environments and solving complex tasks.Therefore,it is crucial to study the motion control of robotic arm systems with humanoid behavior.Previous studies have been conducted to improve the humanoid learning control of robots,but due to limitations such as uncertain grasping posture and insufficient feedback control ability,there is still a lack of robotic arms that can be used for practical complex grasping and control tasks in complex environments.Therefore,it is of great significance to research and develop an arm integrated robot with high integration,low programming load,and superior control performance that mimics human grasping control.Due to the fact that the control of a robotic arm is different from the regulation of the human brain,it is necessary to design and develop effective controllers to accurately control the grasping motion behavior of the robotic arm.The two common types of control methods for grab control are feedforward control and feedback control.Feedforward control relies on identifying the user’s movement intention,selecting an appropriate preset mode,triggering the start of the grasping action,causing the finger joints to move in a predetermined manner.Feedback control obtains external environmental information as feedback input,thereby adjusting the finger joint angle in real-time to adapt to various unknown states.Common external signals that can be used for feedback include visual signals,tactile signals,etc.The existing robotic arms have many shortcomings in both feedforward control and feedback control.For example,the degree of freedom of feedforward control is too high,and the decoding of motion intentions is too complex.It is urgent to reduce the degree of freedom of motion control.Under feedback control,how to more effectively utilize real-time force feedback information and improve the control efficiency of grasping uncertain objects is the key to improving grasping stability.Therefore,it is crucial to establish appropriate feedforward and feedback control mechanisms to improve the humanoid grasping performance of robotic arm.Based on the above considerations,this article constructs a multi degree of freedom robotic arm system from the perspectives of integrating operating systems,reducing task learning costs,and improving control performance.A suitable method for learning humanoid grasping skills is established,and a humanoid grasping control strategy combining feedforward and feedback is designed to achieve highly humanoid grasping operations.This article discusses in detail the work done in system development and integration,as well as learning and control strategies for humanoid grasping skills.The main progress and results achieved are as follows:(1)A system integrating the robotic hand and the mechanical arm has been built,and a data glove with high sensitivity bending detection function has been developed to achieve real-time acquisition of angle data of the human five finger joint.The data can be wirelessly transmitted to the control system,achieving humanoid grasping posture mapping of the robotic hand.A wireless force acquisition device has been developed to obtain fingertip contact force information when grasping objects,and use the contact force as feedback input to the controller.Integrate the development environment of the devices used,design two user operation interfaces,establish functions such as data input,posture selection,and image display,and achieve the basic motion functions of the robotic arm.A water bottle grasping experiment was designed,and through testing the aforementioned humanoid grasping platform,it was found that the constructed multi degree of freedom robotic arm humanoid grasping platform can complete coherent and reliable grasping motion tasks.(2)Develop gripping skills learning methods suitable for mechanical arm.In a physical guided environment,the joint trajectory is obtained by directly dragging the end of the mechanical arm.The trajectory is modeled and learned using Dynamic Motion Primitive(DMP),and then DMP models are established for each of the 7 joints to improve the accuracy of trajectory reproduction.The operating environment under the space demonstration is set up,and the 3D space motion data of the human hand based on the motion capture system and inertial measurement unit(IMU)is established,and it is mapped to the 3D space to form the motion track of the end of the manipulator.The obtained IMU raw data is processed,and the data in the world coordinate system is obtained through coordinate conversion,gravity cancellation,and other technologies.The trajectory accuracy is further improved through motion state recognition,motion speed compensation,and other techniques.Different track results are compared by calculating mean absolute error(MAE),maximum absolute error(MAX),and root mean square error(RMSE).The results indicate that under physical guidance,the MAE,MAX,and RMSE of the learned trajectory and the original trajectory remain on the order of 10-3 or below,indicating that DMP can achieve good learning performance in a single demonstration trajectory and has good generalization ability.The MAE,MAX,and RMSE parameter values obtained by using the method to process IMU data in a spatial demonstration are significantly smaller than those obtained without speed compensation and motion discrimination methods,thus proving the effectiveness of the method.(3)The feedforward and feedback control system of humanoid grasping of the robotic hand is studied and designed.In the feedforward control,33 human hand grasping gesture datasets were first established,and combined with the driving structure of the robotic hand in this study,9 gestures were selected to map to the robotic hand to construct a pre-grasping dataset.Utilizing posture synergy,principal component analysis(PCA)is used to process the pre-grasping dataset,find the collaborative relationship between the robot’s driving joints,and reconstruct the grasping posture.In feedback control,the collected contact force information is used as feedback input,and a force feedback controller is constructed referring to the principle of admittance control to find the relationship between the difference between the target force and the actual force and the position of the joint motor.The finger joints are defined according to functional divisions,and force feedback control is introduced for the joints defined as G(grip).The remaining joints maintain their current positions after generating a pre-pose.The above method is defined as a humanoid grabbing controller,and a baseline controller based on collaboration and pre-programming is set up.The research results show that the sum of the first three principal components in all principal component information processed by 9 grasping poses exceeds 90%,and the average reconstruction error obtained by reconstructing the poses is 6.4%.After using K-means clustering classification,it was found that the dataset can be divided into two categories.To further improve the reliability of data classification,9 new grasping postures were added and PCA analysis was conducted on 18 postures.The results showed that the reconstruction error of 14 postures decreased after classification,and the overall average reconstruction error of the above 18 postures decreased from 6.6%before classification to 4.1%after classification,indicating that data classification can effectively improve the reconstruction accuracy of grasping postures.(4)Finally,the overall test of the humanoid grasping platform is carried out under the two environments of physical guidance and space demonstration.Under the physics guidance,grasp objects of three shapes by setting the original trajectory,changing the starting and ending points,the learning and reproduction effects of DMP on demonstration movements were verified,as well as the role of posture coordination in the grasping process.In the space demonstration,grasping movement data of 6 healthy subjects were collected,the data recorded by the motion capture system and IMU was compared.Comparing the data obtained by the motion capture system with the data collected by the IMU,it was found that the data obtained by the motion capture system was better.Therefore,it was mapped to the three-dimensional space at the end of the robotic arm after DMP learning.Five gripping tests were conducted on 10 objects of different sizes and shapes,with controllers using humanoid gripping controllers and pre-programmed baseline controllers.The results show that DMP can perform well in demonstrating trajectory learning,with a success rate of 84%and a success rate of 42 out of 50 grasping tasks under the humanoid grasping controller,which is significantly higher than the pre-programmed baseline controller.This study aims to meet the development requirements of intelligent and anthropomorphic robotic arms.A multi degree of freedom robotic arm system was constructed,and a suitable learning method for humanoid grasping skills was established.A humanoid grasping control strategy combining feedforward and feedback was designed to achieve highly humanoid grasping operations.It has a positive effect on solving the problems of low system integration,high programming task cost,and poor performance under uncertain grasping targets,laying a solid foundation for building a new generation of arm integrated robot systems with low cost,strong applicability,and superior performance.