Research On Demonstrations And Explorations Based DMP Robot Skill Learning

As a branch of the robotics field,robotic arms are increasingly used in industrial manufacturing.From welding assembly to part handling and processing,robotic arms play an increasingly important role.With the emergence of higher production efficiency requirements,robots are required to become intelligent,and there have been robots to replace human and human-machine collaboration and other higher and more complex technical needs.Robot skills learning systems are also required to be interpretable and autonomous.The personalized production mode requires the robot arm movement trajectory to be easy to express and update,its learning system should be adaptable and self-organizing in the face of change,and its skill model should be reusable in similar tasks.In this paper,according to the robot arm issue,that traditional teaching methods,when switching new tasks can not retain skills,or flexibly adjust and optimize,based on teaching and exploration of DMP skills learning algorithm research,the main research content is as follows:Firstly,a DMP fitting method based on teach-in trajectory is proposed,the end trajectory of N-shaped robot arm is collected by dragging teaching method,and a sample can be trained for DMP model by discrete time differential pre-processing.A mathematical model of dynamic motion substrates for robot complex trajectory learning is established,a training sample is fitted with radial base function fitting algorithm,and complex shape trajectory re-emergence is realized by means of DMP nonlinear terms,which lays the foundation for the optimization of the strategy parameters with the nonlinear item weight parameters of the motion model.Secondly,a policy-search reinforcement learning method based on path integration exploration of DMP model is proposed.The policy optimization algorithm and the target point evaluation function are designed to realize the rapid adaptation of existing skills in the new task scenario.Under the new task after switching the target point,the initialized DMP model of the sampling track is established,its optimization trajectory is generated,and the experimental verification of the algorithm simulation is completed.Finally,in order to verify the effectiveness of the DMP skill learning algorithm based on teaching and exploration,by setting the complex trajectory verification scheme of different shapes,using the URSim robotic arm motion software developed by UR company,the data collection of drag demonstrated trajectory is completed,and the simulation and real experiment research is carried out by using the track pre-processing of discrete time difference,track fitting based on radial base function and track optimization algorithm based on path integral,and the proposed skill learning method is more flexible than the traditional single coding and single teaching.The adaptability and re-emergence of the robotic arm to mission changes is greatly expanded.