Collision free path planning of an object using multiple robot manipulators.

摘要

This thesis presents two algorithms for the path planning of robot manipulators, specifically for multiple manipulator systems attempting to transform a single object without the need for a closed form inverse kinematic solution for a specific manipulator. The first algorithm is a dual quaternion Jacobian based numerical inverse kinematic method. The Jacobian is formulated to transform the change in joint space of the manipulator to the change in dual quaternion transformation space of the manipulator's end effector. Joint space solutions are found that satisfy both the position and orientation of the end effector. The second algorithm is an adaptation of the sampling-based rapidly-exploring dense tree algorithm. The algorithm is adjusted to handle multiple manipulators cooperatively transforming a single object while avoiding environment collisions and invalid joint spaces. The tree generation algorithm forms extensions for the tree by extending the manipulators toward dual quaternion transformations of the object in the environment using the aforementioned numerical inverse kinematics algorithm. This avoids having to find random joint configurations of each manipulator that satisfy the closed link constraints caused by the grasps of each manipulator on the object. The algorithm is able to develop a path from an initial transformation to a goal transformation to the object using multiple robot manipulators. The algorithms presented are tested under a pair of simulation environments while varying the algorithm parameters as well as environment components such as the location of the manipulators as well as the presence of obstacles. Two seven degree of freedom Schunk manipulators were used for the experiments of this thesis.