An Agent-Based Approach to the Design of Rapidly Deployable Fault Tolerant Manipulators

There exists a need for manipulators that are more flexible and reliable than the current fixed configuration manipulators. Indeed, robot manipulators can be easily reprogrammed to perform different tasks, yet the range of tasks that can be performed by a manipulator is limited by its mechanical structure. In remote and hazardous environments, such as a nuclear facility or a space station, the range of tasks that may need to be performed often exceeds the capabilities of a single manipulator. Moreover, it is essential that critical tasks be executed reliably in these environments. To address this need for a more flexible and reliable manipulator, we propose the concept of a rapidly deployable fault tolerant manipulator system. Such a system combines a Reconfigurable Modular Manipulator System (RMMS) with support software for rapid programming, trajectory planning, and control. This allows the user to rapidly configure a fault tolerant manipulator custom-tailored for a given task. This thesis investigates all aspects involved in such a system. It describes an RMMS prototype which consists of seven manipulator modules with a total of four degrees-of-freedom. The reconfigurability of the hardware is made transparent to the user by the supporting control software that automatically adapts itself to the current manipulator configuration. To achieve high reliability, a global fault tolerant trajectory planning algorithm is introduced. This algorithm guarantees that a manipulator can continue its task even when one of the manipulator joints fails and is immobilized. Finally, all these aspects are considered simultaneously in the task based design software, that determines the manipulator configuration, its base position, and the fault tolerant joint space trajectory that are optimally suited to perform a given task. The most important contribution of this thesis is a novel agent-based approach to solve the task based design problem. The approach is based on a genetic algorithm for which the modification and evaluation operations are implemented as autonomous asynchronous agents. Specific design knowledge about the task based design problem has been included in the agents, resulting in a significant reduction of the size of the design space and of the cost of evaluating a candidate design. Furthermore, thanks to their autonomous and asynchronous nature, these agents can be easily executed distributedly on a network of workstations. The flexibility and performance of the agent-based implementation, combined with the problem specific knowledge included in the modification and evaluation agents results in a powerful new approach to task based design of rapidly deployable fault tolerant manipulators. Finally, the thesis presents a performance analysis of the agent-based design framework by comparing its results with those of exhaustive search, random search, and multiple restart statistical hill-climbing. This analysis is performed for three examples, including a comprehensive example of a satellite docking operation with a fault tolerant modular manipulator mounted in the cargo bay of the space shuttle.

Sponsor

Department of Energy (Grant #DE-F902- 89ER14042)
Sandia National Laboratories (Contract #AL–3020)
The Robotics Institute at Carnegie Mellon University

Date

1996-08

Resource Type

Text

Resource Subtype

Dissertation

Full item page