Research On The Motion Planning Method Of Anthropomorphic Arm Based On Human Arm Movement Primitives And Motion Prediction

Human-robot interaction technology is a popular development direction for current and future robotics,involving many fields such as daily life,manufacture,and health care.In human-robot interaction,the anthropomorphic arm,as a key component of the interactive robot to perform complex tasks,interacts most closely with humans.To ensure the efficiency and safety of the interaction process,improving its intelligence level and the capability of imitating human motion is a significant problem that needs to be solved.This paper focuses on how to make the anthropomorphic arm generate intelligent and human-like interactive motion and uses human arm movement primitives and human arm motion prediction as the technologies to study the method of human-like motion planning of the anthropomorphic arm in human-robot interaction.The main research contents are as follows:(1)The research on the theoretical system of human arm movement primitives.A new human arm movement primitive model is established,and its basic usage rules are developed to solve the problem of expressing and modeling the motion of the human arm or anthropomorphic arm in an intuitive,efficient and comprehensive way.First,the structure of the human arm and the characteristics of its motion generation are analyzed,and the kinematic model of the human arm is studied.Second,the commonalities of existing movement primitive models are investigated,and the modeling principles of human arm movement primitives are established based on the application requirements of them in human-robot interaction.Then,according to the proposed modeling principles,the model of human arm movement primitives is established,and a library of human-arm movement primitives is built,and the general rules for connecting the primitives and quantifying its parameters are developed.Finally,the recognition method of human-arm movement primitives is established,and an example investigation of human-arm movement primitives is conducted based on motion capture experiments.(2)The research on the mutual mapping method between human arm movement primitives and the anthropomorphic arm's joint angle.A generalized method for the mutual mapping of human arm movement primitives and the anthropomorphic arm's joint angle is proposed to solve the problem of fast transition between movement primitives and joint angles.Firstly,the structural characteristics of the anthropomorphic arm are presented,and its kinematic model is established with KUKA iiwa as an example.Second,a general method of mapping human arm movement primitives to anthropomorphic arm joint angles is proposed to realize the motion control of anthropomorphic arms using movement primitives.Then,the general method of mapping the robotic arm's joint angle to the movement primitive is proposed,realizing the motion modeling of the robotic arm using the movement primitive.Finally,the proposed mapping method is validated by implementing experiments on motion control and motion modeling of anthropomorphic arms based on human arm movement primitives.(3)The research on task-motion planning method based on human arm movement primitives.A task-motion planning method based on human arm movement primitives is proposed to achieve autonomous task planning and motion planning for anthropomorphic arms.Firstly,the task-motion planning method framework is established,which is divided into task layer,primitive layer,and joint layer.Secondly,a STRIPS language-based autonomous task planning method and a minimum energy consumption index are established,which achieves the segmentation of complex tasks and the optimal subtask sequencing.Then,a motion planning method based on the selection,ranking,and quantification of movement primitives is proposed,and the primitive laws in human arm motion are transferred to the planning method.Finally,the robot experiment of stacking cubes is implemented to validate the proposed taskmotion planning method.(4)The research on human arm motion prediction method.A new human arm motion prediction method based on the human arm motion model and machine learning model is proposed to predict the time,endpoint,and motion trajectory of the human arm reaching movement.First,numerous human arm reaching movement capture experiments are implemented,and a human arm reaching motion database is established.Secondly,based on the motion database and the minimum jerk model theory,a modified minimum jerk model is established to realize motion trajectory prediction.Then,based on theoretical analysis and experimental data validation,the optimal start time for motion prediction was determined.Then,four Gaussian process regression models were built and trained for predicting the endpoint and total duration of the motion.Finally,simulation validation and experiments of human armanthropomorphic arm position docking interaction were implemented to verify the proposed motion prediction method.The motion prediction results provide the referenceable motion destination and time for motion planning.(5)The research on human-like motion planning method of the anthropomorphic arm.A human-like motion planning method based on human arm motion patterns is proposed to generate high-level human-like motion for anthropomorphic arms in human-robot interaction.Firstly,the effects of the motion region segmenting and motion phase segmenting on the occurrence probability of human arm movement primitives and the connecting mode of primitives were studied.Secondly,a motion pattern extraction algorithm is established to extract the motion pattern of the human arm,and the distribution of each motion pattern in human arm motion is analyzed.Then,a new motion planning method considering the human-like motion pattern is established,a pattern recognition neural network-based motion pattern determinator with three human-like levels is designed,a motion time designer based on Fitts law is established,and a joint angle solver based on backpropagation neural network is proposed.Finally,a human-like motion planning APP and V-Rep experimental platform are developed to validate the proposed anthropomorphic arm motion planning method.