Research On Minimum-Time Trajectory Planning For Robot Point To Point Task

In order to shorten robot's cycle time during its working, the concept of the time optimal trajectory planning is raised, which is a popular research pot among the problems related to robot.At present, when handling the min-time optimal trajectory planning problem of a robot, an interpolation function is usually used to connect each target points, which contains more undetermined coefficients than the boundary conditions can determine. In this way, all the undetermined coefficients that matching boundary conditions form an optimization space and an optimal trajectory will be get through a certain kind of optimization method when set the run time as the objective function. According to this method, the use of interpolation function and optimization method is the key point that has a grate influence on the optimization effect. Concerning this issue, the following studies are carried out:At present time, there is no effective decision theorem to judge a trajectory is time shortest or not in an optimal trajectory planning problem. The concept of "the trajectory equivalent on the problem of point to point" is raised. From the perspective of adequacy, a theorem that cubic spline function can describe the shortest path when the number of its segments between 2 target points is 7 is proved with the idea of equivalent thought.Based on the interpolation function above, an objective function focus on the cycle time is formed. The conjugate gradient method, quasi Newton method and genetic algorithm are respectively used to optimate the objective function and their optimization performance is compared. At the same time, the difference between the results of the optimizations with the interpolation function used in normal way and the interpolation function which is a cubic spline with the segments number between 2 target points is 7 is also compared.Aiming at the problem of redundancy, the objective function is modified and the new objective function is optimated by genetic algorithm, the optimization performance is given.There is some error between the robot kinematics model and the entity in application. To lick this problem, a BP neural network is used to compensating the error. The compensation value is added to the objective function. Finally, the optimization result is loaded on a robot and operated to verify the correctness of the method.