Development for Space Robotics with Emphasis on Fault-Tolerance
with Del Tesar and Dev Sreevijayan

Introduction - The goal is to develop and test technology applicable to all future missions of NASA (lunar base, Mars exploration, planetary surveillance, space station, etc.). This technology would be in balance with the astronaut sharing tasks based on performance, cost, and availability issues. In order to reduce costs, the system would be made up of a finite number of modules (both hardware and software) proven by extensive testing in space. This set of modules would be constantly under technical development so that "tech mods" would be feasible at any time. Also, the repair and logistics functions (warehousing of spares in space) would be based on these modules to further reduce costs. This architecture would allow the specification of a robot configuration "on demand" reducing the threat of obsolescence and freeing the mission planner to aggressively use advanced (yet proven) technology. The following are some of the technologies required for fault-tolerant space robotics:

 1. Actuator Technology - Present actuator technology is largely unchanged since 1965 except for the utilization of rare earth motors and improved electronic controllers. The goal is to aggressively develop component technology which can be integrated in a carefully designed class of actuator modules made up of dual motors, brakes, gear drives, clutches, sensors, electronic controllers, etc., which would provide fault tolerance for dramatically improved performance and reliability of space mechanisms including robotics.

2. Modular Architecture - A true modular architecture (in the same form as has proven useful for computer systems) can not only reduce life cycle costs (repair, tech mods, logistics spares planning, etc.) but can dramatically increase performance while allowing the designer to more freely and quickly develop actual operating systems to satisfy future space missions. It is proposed to assemble and reconfigure a broad population of systems from a very small collection of proven and optimized modules produced at lower costs.

3. Task Planning - The complex motion of a body in space to trace out precision trajectories requires the sophisticated theory of algebraic curves to smoothly coordinate all 6 DOF of the end-effectors. the need for dependable task planning derives from a spectrum of demanding physical tasks such as debris damage inspection, precision wiring disassembly and assembly, force fit assembly, dual arm operations, etc. while avoiding obstacles. The goal would be to make task planning more automated requiring primarily supervisory involvement by the astronaut reducing their time burden and potential fatigue.

4. Dual Arm Operations - Due to the lack of frictional stability generated by gravity forces, all parts must be under control at all times to prevent "dropping". This means that either special fixtures (the bane of data base control in manufacturing) must be employed or dual arms must perform the relative motion tasks (force f@t assembly, control of ungainly objects that may be easily damaged, removal of insulation wrappings, bending to fit,'etc.) that are sure to occur on long duration missions. No real time operation of a dual arm system capable of these tasks exists today. For two manipulators of 7 DOF each, this requires a level of control (precision force and position of 14 inputs to control 6 relative outputs) far beyond any standard approaches (PID, fuzzy logic, sliding mode control, adaptive control, etc.).

5. Task Performance - Long duration space missions suggest an enormous range of physical tasks of great complexity (handling of large modules, precision welding and forming, unstructured tasks associated with joining and fastening, precision machining, etc.). This complexity can be met only by a criteria based decision control structure based on accurate system parameters (using careful metrology) and hundreds of performance criteria. A prioritized selection of these criteria will be used to create performance indexes to compare the model based performance with the actual performance derived from a very broad collection of sensor signals. Differences between actual and modeled performance will be the basis for adjusting the control inputs to the system.

6. Condition Based Maintenance - Having established a model reference control structure comparing actual with predicted performance, it becomes feasible to monitor the system over time to determine when basic maintenance (replacement of actuator components, sensors, controllers, etc.) should be performed and to provide an archival record of that performance. This should improve the system's reliability, reduce the cost of operation, prevent unexpected failures, and provide lesson's learned for the operator and the designer of future components as well as to the mission planner for module selection to make up systems for other tasks.

7. Fault Tolerance - Fault tolerance is virtually non-existent in present robotics development for space. A full architecture for fault tolerance involves four levels (alternate physical pathways) of mechanical structure to avoid faults. The UT program strongly recommends a 1 0 DOF manipulator system (level III) made up of dual actuators (level 1).. This level of choice (20 actuator inputs to control 6 outputs) can only be achieved by a criteria based decision making structure based on performance indexes composed of hundreds of physical criteria (which demands an extremely high computational capacity). Such superior system controller technology (several gigaflops) is emerging as a commodity (at reasonable cost) in the near term. Hence, fault tolerance is not only feasible but R can only be achieved through a comparative analysis between an accurate and complete analytical model reference and a sensor based actual model of the system. This makes Fault Detection and Isolation (FDI) possible. No other method of control does.

8. Man-Machine Interface - Because of the extraordinary value associated with the time of the astronaut, the interface between man and machine is being recognized as a key resource to maximize overall performance and to train (skill) the system's operator. Very complex operations (dual arms, disturbance rejection, unstructured tasks, precision assembly at small scales, multiple slaves, obstacle avoidance, etc.) require an exceptional level of dexterity and task performance. This is best achieved by setting operational priorities (selection of criteria, performance indexes, threshold levels for fault identification, etc.) by human intervention. Specially designed actuators, human augmentation software, fault tolerance, etc., must be built into future manual controllers to maximize the task performance of an increasingly complex slave manipulator technology.

9. Ground Based Control - As space missions develop (by analog, with the aircraft pilot), the astronaut will be less available to perform mundane, repetitive, and low valued tasks. In order to reduce costs, the demand on the astronaut's time, and to reduce risks, the robot will either have to be operated remotely from a protected module or it will have to be operated from a stand-off position (say the moon or from a control center on earth). This set of conditions leads to the inevitable conclusion that an enhanced man-machine interface to remotely control an array of deployed slave manipulators (robots) in space is essential. Next Page ->