Research On Motion Planning Algorithm Of Manipulator Based On RRT And Dynamic Motion Primitive

Motion planning is an important part of the research on manipulator,and planning algorithm which based on Rapidly-exploring Random Trees(RRT)is often used for path planning in the configuration space of manipulator.Most of these algorithms feature randomness and the paths planned by them need extra time to optimize.Aiming at this problem,an improved Rapidly-exploring Random Trees Star Fixed Nodes(RRT*FN)path planning algorithm is proposed in this thesis,which can save planning time while planning a better path.Meanwhile,in most cases,the motion planning of manipulator is only served with a specific task,but when a new task is given to the manipulator,it needs to be re-planned.In view of this,this thesis uses dynamic motion primitive algorithm to learn the trajectory planned by the improved RRT*FN algorithm for the generalization of the motion trajectory.The main points of this thesis is as follows:(1)This thesis takes UR5 manipulator as the research object.Firstly,it carries out the kinematics analysis of UR5 manipulator,calculates the forward kinematics of UR5 manipulator by the improved DH parametric method,calculates inverse kinematics of UR5 manipulator through geometric method,and analyzes singularity of UR5 manipulator.(2)For the randomness of most planning algorithms which based on RRT,and the need for extra time to optimize the paths planned by them,an improved algorithm of RRT*FN is put forward in this thesis.By using a heuristic sampling method,combining hyper-ellipsoid subset area and the previously obtained path information,this algorithm can simple more purposely.An improved node deletion method is adopted,that is,different weights are given to leaf nodes both within and outside of the sampling area.Therefore,this algorithm can retain more valuable leaf node information,and keep the optimization characteristics while accelerating running speed.In addition,in the process of sampling nodes,this thesis uses bounding volume hierarchies and sweep and prune algorithm to solve the collision detection problem of the manipulator.After the path is planned in the configuration space of manipulator by using the improved algorithm,the combination of path time parameterization can make planned path have speed information,so as to execute motion trajectory on the real manipulator.(3)As most motion planning of manipulator is only served with a specific task,this thesis takes the trajectory planned by the improved RRT*FN algorithm as the demonstration trajectory,and uses bio-inspired dynamic motion primitive algorithm to learn it.When the task of the manipulator is changed,the generalization of dynamic motion primitive algorithm can be used to generate a motion trajectory which is similar to the demonstration trajectory and meets the need of task,by which to improve the applicability of manipulator in various tasks.