This thesis presents a study of the optimal design of a class of six degree-offreedom (DOF) closed-chain manipulators. This class of manipulators, which are t rmed hybrid manipulators, consist of serial branches, each comprised of actuated and passive joints, acting in parallel on a common end effector. Dexterity measures based on instantaneous kinematic characteristics of the manipulator are used as the primary obj ective in isolating optimum designs.

The fully-parallel Stewart platform, which represents a limiting case of a hybrid manipulator where only one joint in each branch is actuated, is first xamined. As an initial design step, manipulator configurations (manipulator architectures and end effector positions and orientations) optimizing local dexterity are determined. For a platform centred reference location and a given length for scaling purposes, a twoparameter family of optimal configurations is shown to exist. Through the use of a new performanc measure based upon gradients of local d xterity measures, a unique optimum Stewart platform architecture is isolated from those po essing optimum local dexterity. The resulting optimum manipulator architecture is one in which the dimensions of the base are twice those of the platform and the linear actuator attachment points at the base and platform meet in alternating pairs.

Hybrid manipulators are then examined. Through consideration of preferred attributes relating to the performance of th manipulator, a specific hybrid chain structure is selected from possible six DOF structures for further investigation. A class of kinematically simple serial-chain branches suitable for the chosen hybrid structure is defined and arguments based upon kinematic equivalency and mobility are used to show that only five unique branch structures with revolute joints belong to the kinematically simple class. A novel approach to manipulator configuration optimization for optimal local dexterity objectives is introduced. This new approach involves finding geometric characteristics of manipulator configurations which optimize dexterity and then finding actual manipulator configurations fitting these characteristics. The method is applied to find optimal configurations of hybrid manipulators utilizing the previously identified branch structures.