Learning Hybrid Force Position Skills From Demonstration For Robotic Assembly Tasks

Assembly is the task of joining together a group of parts.Some processes require pushing,stamping,and twisting behaviors,which can generate huge contact forces,and other processes may require precise positioning.Therefore,position and force must be properly controlled to complete the assembly task.Force-based assembly solutions are often designed for specific tasks and configurations.The design,programming and debugging process is time consuming.Learning from Demonstration(LfD)is an effective way to implicitly transform human knowledge into robots.It relies on a demonstration method to guide the robot and record sensor data at the same time,learn skills and perform task reproduction from the data recorded during the demonstration.The existing demonstration technologies have their own application scenarios and limitations.Gesture-based methods provide an unconstrained natural way to operate a robot.However,the uncertain and low-resolution features of human motion and the lack of force feedback make it impossible to produce compliant and precise robot motion.This paper presents a novel gesture-based flexible assembly skill demonstration system.By using the left hand as the commander and the right hand as the positioning,you can instantly adjust different operating modes and zoom ratios to meet accuracy and efficiency requirements.In addition,a vibration-based force feedback system was developed to provide the operator with a remote telepresence to sense the contact force during assembly.Demonstrating learning based on a certain skill alone is inefficient and cuts off its contextual connection during assembly.Skills are interconnected with pre-skills and postskills,with hidden boundary conditions and constraints.Skills,environments,and tasks also contain multiple constraints.These constraints can more accurately reflect the knowledge behind the skills than the motion trajectory.Therefore,this paper proposes a compliant assembly skill learning framework based on task constraints,and studies task division,control strategy selection,and skill modeling and task reproduction methods in compliant assembly.In addition,in the process of task reproduction,when the environmental parameters change,in order to ensure the safety of the robot and its components,a robot linkage obstacle avoidance algorithm is proposed in this paper.Finally,the peg-in-hole assembly task was used to test the effectiveness of the demonstration system presented in this paper.In order to verify the flexible assembly skill learning method proposed in this paper,a high-precision peg-in-hole assembly experiment based on Novint Falcon platform was performed,which proved the effectiveness of the algorithm for this skill learning.Later,two experiments were performed on two remote operation platforms based on gestures and Novint Falcon.It is concluded that the gesturebased teleoperation system proposed in this paper can perform sub-high-precision operations due to its good scalability,and can choose a suitable working mode to make the sub-high-precision tasks complete faster.Finally,an adaptive obstacle avoidance experiment is carried out on a gesture-based demonstration platform,which verifies the effectiveness of the robot linkage obstacle avoidance algorithm proposed in this paper.