MOTION PLANNING FOR MANIPULATORS: A NOVEL APPROACH WITHIN A SEQUENTIAL FRAMEWORK
Xinyu Zhu · Summit (Simon Fraser University) · 1994
This thesis presents a novel approach within a sequential framework to develop practical motion planners for manipulators with many degrees of freedom (DOFs).The sequential framework is a paradigm for developing motion planners (Gupta 1990; Gup'ia.and Guo 1994).The essence of the sequential framework is to exploit the serial structure of m a ~i p ulators and decompose the n-dimensional problem of planning collision-free motions for an n-DOF manipulator into a sequence of smaller (mj + 1)-dimensional sub-problems, ml + mz 4.-.. .t ms = n.The jth sub-problem corresponds to planning the motion of a sub-group of m j links (DOFs) along a given path.Backtracking is used if no path exists for a given sub-problem.In this thesis, each sub-problem, a 2-dimensional motion planning problem, is solved by using numerical potential fields defined over bitmap-based representations.Furthermore, a systematic and efficient backtracking mechanism based on a novel notion of virtual forbidden regions in subspaces is presented.This approach leads to more efficient and robust motion planners than a previous implementation that used polygonal representations and visibility graphs (Gupta and Guo 1992).The new approach has been successfully applied in both 2 and 3 dimensional workspaces.Experiments have been conducted for manipulators with up to 8 degrees of freedom among randomly or manually placed obstacles.Although it is not complete, the planner never failed for these examples in hundreds of simulations, and at most three backtracking levels were needed.For a 7-DOF manipulator in 3dimensional workspaces, the average run-time is about 8 minutes.These empirical results clearly show that the approach is efficient and practically useful.Guo, Hezhi Fan and others for their vduable help and discussions on my thesis.