Research On Skill Learning And Control Methods For Robotic Manipulation

As an important support of the new generation of artificial intelligence,robot manipulation technology has been widely applied in different areas.However,most of the current robots are deployed and restricted into a structured environment,where robots can only perform some single,repetitive,and regular tasks by using the pre-programming strategy.Recently,artificial intelligence as an important approach to improve the intelligence level of robot manipulation has been integrated into robot technology to realize the understanding,learning and control of the robot for operation skills without tedious manual programming and debugging.Human-robot skill transfer technology can quickly transfer human compliant manipulation skills to robots by skill learning and compliance control methods,and improve the flexibility of robot manipulation,which has become a hot spot in the research field of robot manipulation.Based on the above research background,this paper studies the skill learning method and intelligent control technology for robot manipulation,which aims to enable robots to perceive human's motion intention and environmental information independently through human-robot interaction system,imitate human's operation action to directly learn human's experiences and skills,and finally realize the robot's safe,stable and flexible manipulation through compliance control methods.The main works and contributions of this paper are summarized as follows:(1)Two kinds of human-robot skill learning interaction systems are designed for robot manipulation: on the one hand,a human-brain-robot neural interaction system based on the steady-state visual evoked potential is designed,where the robot can perceive human brain intentions and then generate advanced neural control commands;on the other hand,a physical interaction system based on tactile force information is designed,where the robot can use the force sensing information and the admittance model to learn human's compliance trajectories.(2)For the problems of navigation and obstacle avoidance of mobile robots in complex environment,a brain-robot cooperative navigation and obstacle avoidance strategy based on multi-information fusion is proposed.Firstly,a simultaneous localization and mapping method based on the particle filter multi-information fusion is designed to ensure the robot's real-time positioning and accurate mapping in the indoor environment with multiple obstacles.Secondly,an artificial potential field based on brain intention perception and learning is designed for obstacle avoidance path planning,which uses brain intention to change the distribution of artificial potential field,establishes the control relationship between brain intention and the distribution of artificial potential field,and makes real-time path planning in multi-obstacle environment.Then,a trajectory tracking controller based on Lyapunov method is designed to ensure the smooth operation of the mobile robot with nonholonomic constraints.Finally,the effectiveness of this brain-computer cooperative navigation and obstacle avoidance strategy is verified in the porch environment with multiple obstacles.(3)Aiming at the humanoid manipulation of redundant exoskeleton arm,a closed-loop bionic control strategy based on brain intention perception and learning is proposed,where the neural signals can directly control the exoskeleton arm to complete the fine manipulation task.Firstly,the human-brain-robot neural interaction system is used to realize brain intention perception and recognition.Secondly,a bionic neural control method in polar coordinates of task space is proposed,which maps brain intention directly to the movement of the end-effector of exoskeleton arm in polar coordinates,and then realizes the motion planning of exoskeleton arm in operation space.Thirdly,a primal dual neural network optimization method based on linear variational inequality is applied to transform the trajectory in task space into the trajectory in joint space to obtain the redundant optimal solution in joint space,thereby avoiding the problems of matrix inversion and joint singularity.In addition,an adaptive joint space controller is designed based on robot dynamics.The tangent barrier Lyapunov function is designed to solve the state constraint problem,meanwhile,a disturbance observer is introduced to estimate the external disturbance and compensate for the uncertainty and approximate tracking error of the system.Finally,the effectiveness of the closed-loop bionic control strategy is verified in the manipulation task of redundant exoskeleton arm.(4)To adress the problem of human-robot cooperation with touch as physical interaction information,a hierarchical human-robot cooperative compliance control strategy is proposed.The proposed compliance control strategy consists of two levels,i.e.,high level and lower level.The high level is the skill learning strategy,which uses the dynamic motion primitive and Gaussian mixture model to model the robot motion and learn human manipulation skills from multiple demonstrations.The lower level is the compliance control strategy for human-robot cooperation,which ensures the compliance movement of exoskeleton robot during cooperative manipulation by combining admittance controller and joint motion controller based on robotic dynamics.Here,the admittance controller obtains human's compliant manipulation trajectory through tactile interaction information and then transmits it to the end-effector of exoskeleton robot.The joint motion controller based on robotic dynamics adopts the integral barrier Lyapunov function based adaptive neural controller(IBLF-based NN)to track the manipulation trajectory learned by the robot and ensure the robot to move naturally and smoothly like human during human-robot cooperation.In addition,the proposed low-level compliance control strategy solves some problems of the dynamics control of exoskeleton robot system.For example,we design an integral barrier Lyapunov function to solve the state constraint problem and suppress the bad vibration problem of exoskeleton robot system;a radial basis function neural network is utilized to estimate the uncertain part of the dynamics model;and the disturbance observer is introduced to compensate the approximate residual error of the neural network,the input nonlinear disturbance part and the external disturbance,so as to further improve the robustness of the control system.Finally,the effectiveness of the compliance control method is verified in the human-exoskeleton robot collaborative task.