Development for Space Robotics with Emphasis on Fault-Tolerance, page 2

Architectures - The level of performance and versatility expected from space robots makes ft imperative that particular emphasis be placed on the question of architecture across multiple domains. While such design issues have to be addressed for each of those domains (mechanical, electronic, and software), certain essential principles may be allowed to pervade the system's architectural considerations. These principles are 1) modularity and 2) redundancy.

Traditionally, mechanical systems have had monolithic architectures that do not permit easy repair and replacement, nor the fluent incorporation of advances in component technologies. Ease of repair and replacement are directly linked to the availabiliy of the system and is, therefore, a matter of immediate concern to space systems. The deficiencies of such a philosophy, or lack of one, are best offset by aiming for an architecture that is highly structured and modular. A true modular architecture helps reduce life cycle costs and frees the designer to quickly prototype and develop actual operational systems for future space missions.

Redundancy is demanded by twin operational considerations for space operations: safe and enhanced performance provided by redundant systems--to be explained in a later section--and the ability to tolerate faults. Fault tolerance in a robotics context may be defined as the capability of the system to sustain a failure and still continue operation without significant impact on the manipulator payload or its immediate environment. Graceful degradation is often inadequate as a requirement and uncontrolled motion must at all costs be minimized. Fault tolerance is assured by the incorporation of protective redundancies and their organization into an effective and responsive architecture. While the causes of unreliability do not disappear, their effects are counteracted by the capably of the system to be intelligent, and to mask the failure or reconfigure itself in the event of component failure. It must be emphasized that these redundancies are active redundancies that are operational at all times and which enhance the performance of the robot through the optimization of secondary criteria. We now consider general modular manipulator architectures for fault tolerance.

The Four-level Architecture - While the issue of architecture depends ultimately on the context and specific tasks envisaged of the robot, ft is possible to devise a broad architectural scheme that is based on the requirement that the system be two-fault tolerant. The resulting architecture, in its most general form, is capable of providing a masking redundancy at the first level and a dynamic or reconfiguring redundancy to tolerate the second fault. The organization of these redundancies may be conceived in four levels and they constitute a subsumptive architecture. The four levels are: 1) extra actuators per joint (e.g., prime mover duality), 2) extra joints per DOF (redundantly actuated parallel structures, e.g., 4-legged spherical shoulder), 3) extra DOF per arm (redundant manipulators), 4) extra arms per manipulator system (e.g., dual arm robots, four-fingered hands, etc.).

2-DOF Redundant Knuckle Mechanism - The knuckle can operate either as a force feedback joystick or as a 2 Degree-Of-Freedom (DOF) manipulator. The knuckle demonstrates modularity and Level I fault tolerance at the servo level. It uses two independent servo systems per single DOF to obtain Level I mechanical redundancy. Modularity is demonstrated in the servo control hierarchy (see description of DISCS). The system is designed to handle a minimum of 1 fault before failing. The system controller acts as a supervisor in analyzing the sensory feedback with a Fault A servo system can either remove itself from the system or be removed from the control hierarchy by its parallel controller. Each servo system consists of a clutch, a brushless resolver, a brake, a Hall-effect sensor and a three-phase Brushless DC motor The system controller consists of a 486 PC operating under the Lynx O/S real-time operating system. The system controller communicates to the servo controllers via a HDLC medium at a rate of 1 MBit/s. HDLC is a communications protocol based on IEEE RS-532 standard.

Modular Brakes for the Fault-Tolerant Actuator - A new brake was developed for use in the next generation of robot actuator technology. The caliper disk brake has an annular geometry that capitalizes on the structural integrity of the robot actuator shell. One to four independent brake calipers can be positioned around the rim of the disk to give a redundant capability. The torque path goes directly from the calipers to the actuator shell to produce a truly, lightweight, integrated design. An uftra-low power consumption is achieved since only a very small current is required to hold the brake pads in the released position. The brake design is very compact, lightweight and has a high torque/weight ratio. Performance parameters were determined from prototype testing and compared to a set of average performance parameters derived from a database of commercially available high performance brake modules. The new brake design has achieved a number of improvements when compared to standard practice.

Object-Oriented Software Architecture - This software provides abstractions for various robotic concepts like kinematics, dynamics, decision making, and fauft-tolerance. The abstractions are implemented as a part of an inheritance hierarchy. This software is developed in a modular and extensible fashion, with the user having the capability of connecting various modules as desired. This promotes rapid prototyping. Also, the framework provides for reuse and extensibility. For example, if a generalized inverse kinematies module would not satisfy user needs, the user could easily substitute that module with a custom inverse kinematics module. Such a change does not affect the structure of the rest of the software. This software provides an ideal environment for robotics research and allows for interaction at all levels of abstraction. Moreover, the rapid-prototyping capability of the software allows for easy experimentation. In addition, due to the general nature of this software, it Is applicable to a wide variety of robotic structures. Next Page ->