Task based kinematic design of robot manipulators

Abstract

In this dissertation, a design methodology called Task Based Design is proposed to design an optimal manipulator for a given task. One motivation of this research is the CMU Reconfigurable Modular Manipulator System (RMMS) that utilizes a stock of assemblable joint and link modules of different size and performance specifications. The modularity in mechanical, electrical and electronic design in RMMS allows the user to design the optimal manipulator for the task at hand. For fully exploiting this feature of the RMMS, it requires development of a strategy called Task Based Design that maps a given task onto a manipulator. The goal of Task Based Design is to design an optimal manipulator which performs a given task best. However, Task Based Design of RMMS as well as general manipulators involves highly nonlinear and implicit functions and a large number of design variables that increase as the given task becomes more complex. To overcome this complexity of Task Based Design, we propose Progressive Design (a high level framework), which decomposes the complexity of the task into three steps: Kinematic Design, Planning and Kinematic Control. Assuming that a given trajectory can be approximated by a finite number of task points, Kinematic Design finds an optimal manipulator and its optimal base position and optimal poses at task points. Planning optimizes poses of the designed manipulator at sub-task points between task points. Kinematic Control connects poses at all task points including the sub-task points, along the given trajectory. Each step requires a low level framework to decompose the complexity of design. Among the three steps, the framework of Kinematic Design is the most important and most comprehensive, and consists of three modules: DOF section, type generation, and an optimization algorithm, and three inputs: task specification, manipulator specification and dexterity measure. We have applied the framework of Task Based Design to three design problems: crank turning task, three straight line motions and space shuttle tile servicing. These three examples demonstrate that our design strategy and framework is effective and efficient for both simple and complex tasks.