| Learning from demonstration has significant advantages such as simplicity,naturalness,and the elimination of manual programming.It is an efficient method for robots to learn operational skills.By learning motion behaviors and variable impedance characteristics from demonstration,the robotic arm can interact closely with humans in a safe and natural manner.Therefore,learning from demonstration of robotic arm based on skeletal point information and variable impedance action generalization are crucial aspects in human-robot interaction,with both theoretical and practical significance.Taking a self-designed humanoid dual-arm robot as the research object,the following work has been carried out: achieving real-time humanrobot joint mapping calculation in the case of arm heterogeneity;identifying highprecision dynamic parameters to obtain accurate feedforward torques;learning motion behaviors and variable impedance characteristics from multiple sets of demonstration trajectories and performing impedance control with variable stiffness generalization based on different tasks.The main research contents include:1.In response to the heterogeneity between the self-designed humanoid robotic and human arm,we have proposed a novel inverse kinematics solution method that combines the screw theory with the D-H method,and successfully achieved humanrobot joint mapping.The heterogeneity issue primarily arises from the inclination angle between the rotation axes of the first joint at the shoulder joint of the robotic and human arm,as well as the offset at the robotic arm’s elbow joint.Furthermore,in scenarios where only skeletal point position information is available from Kinect,without providing posture information,a single inverse kinematics method is insufficient to compute closed-form solutions for all joints.Therefore,in this study,we first employed the screw theory to obtain closed-form solutions for the elbow joint with specific geometric characteristics,and then utilized the D-H method to compute closed-form solutions for the remaining joints.Additionally,to meet the real-time teaching requirements of different teacher,we dynamically adjusted the length of the teacher’s and robotic arm in real-time,allowing the robotic arm to mimic the motion of the human arm in real-time and maintain a similar operational workspace.Compared to other methods,the proposed approach has significantly improved the end-effector trajectory tracking accuracy with increases of 37.36%,21.89%,and73.6% in the x-axis,y-axis,and z-axis directions,respectively.2.To achieve compliance and safe variable impedance control of the robotic arm,an improved particle swarm optimization method for step-by-step parameter identification is proposed to address the issue of accurately calculating the feedforward torque using dynamic parameters.Firstly,several factors in the traditional particle swarm optimization algorithm,including particle initialization,inertia weight,learning factor,and particle mutation,are improved,and the method is utilized for dynamic parameter identification.Compared to other methods,based on the RMS value of each joint torque,the identified parameters using this method improve the torque accuracy of joint 1 to joint 6 by 14.68%,23.0%,9.16%,42.3%,3.73%,and 22.55% or more,respectively.Secondly,an excitation trajectory combining Fourier series and natural logarithm is designed.This trajectory fully stimulates the dynamic characteristics of the robotic arm while avoiding significant vibrations and impacts during motion.Finally,the step-by-step parameter identification method is employed to reduce the dimensionality of the improved particle swarm optimization algorithm during the parameter identification process,thereby enhancing its stability.3.To achieve the transfer of variable impedance skills from the human to robotic arm,a variable impedance control strategy is proposed to learn impedance characteristics from multiple sets of demonstration trajectories.Firstly,the demonstration trajectories are aligned on the time axis through dynamic time warping.Then,using the variance obtained from Gaussian mixture regression,the control strategy represents the variance as impedance control parameters based on an improved Softplus function,enabling variable impedance control of the robotic arm.Compared to other impedance methods,the motion accuracy of joint 1 to joint 4 in the joint trajectories improved by 57.23%,3.66%,5.36%,and 20.16%,respectively.The position accuracy in the x-axis,y-axis,and z-axis of the operational workspace also improved by 46.76%,78.01%,and 33.72%,respectively.This impedance control strategy endows the robotic arm with the ability to withstand external impacts,meeting the requirements of smoothness and safety in human-robot interaction.4.To adapt variable impedance control to different task environments,a segmented generalized approach for variable stiffness based on dynamic motion primitives is proposed.This method utilizes inflection points with monotonicity in the variable stiffness profile as segmentation points and performs segmented generalization to satisfy the requirements of variable impedance control.When the joint trajectories change,the motion accuracy of joint 1 and joint 2 improved by35.14% and 6.04%,respectively,through stiffness generalization.The position accuracy in the x-axis,y-axis,and z-axis of the operational workspace also improved by 10.81%,55.67%,and 11.81%,respectively.When the robotic arm is subjected to significant loads,variable stiffness generalization effectively eliminates deviations in the motion trajectory.Experimental results demonstrate the stability,applicability,and good learning capability of this method.In summary,this paper first utilizes the proposed human-robot joint mapping method to achieve real-time demonstration functionality and learn motion behaviors and variable impedance characteristics from demonstration trajectories.Then,by employing the proposed improved parameter identification method,accurate feedforward torque is provided for variable impedance control.Finally,the proposed variable impedance control strategy and variable stiffness generalization method are adopted to meet the requirements of compliance,safety,and accuracy for the robotic arm in different task scenarios. |
