Robot Path Planning Based On Demonstration

With the development of science and technology,robots are playing an increasingly important role in many fields such as medical treatment,education and service.The research of robot technology is facing great challenges and opportunities.Today,with the rapid development of artificial intelligence technology,robots have been equipped with the abilities of perception,decision-making and memory,which are closer and closer to the concept of "human" in people's imagination.The trajectory planning of robot is the foundation for robot to be intelligent.Robotic trajectory planning aims to compute a collision-free path for the given start and goal configurations.Fast and efficient trajectory planning algorithms are very crucial for robotic system.At the same time,in order to better integrate the robot into human life,robot trajectory is required to be more natural and more in line with human movement habits.Learning from demonstration(LfD)is a promising way to guide the robot behavior.LfD can effectively transfer human experience to robots with the help of human movement characteristics,making robots more autonomous and more anthropomorphicIn this paper,a trajectory planning algorithm of robot arm based on behavior demonstration is proposed,and a method transfers the human motion demonstration is to robot.Based on virtual reality(VR)technology,a virtual environment is built to collect the movement track of human hands.Using the demonstration data,the long short term memory recurrent neural network was trained to generate the trajectory sequence end-to-end.At the same time,the position of human hand was mapped to the end of the robot arm to transfer the trajectory to robot.In addition,based on the demonstration data set,this paper designs and realizes the human-robot real-time interactive grasping system to realize the human-robot interaction in grasping behavior.The main contributions of this paper are as follows1.Based on virtual reality technology and data gloves,a variety of virtual objects are used to guide human beings to grasp.Then a large number of trajectories demonstrated by human hands are collected,analyzed and processed2.Using the learning method and based on the recurrent neural network,the trajectory planning algorithm is designed and applied to the trajectory planning problem of the robot manipulator.The feasibility of the algorithm is verified in the virtual simulation environment3.Design and implement man-machine real-time interaction system.The grasp method of the known model is obtained through human grasp demonstration.Build a visual servo system to identify objects in real time,estimate the pose of objects,so that the robot can choose the right grasp point to grab objects.