The functional requirements for an automated SMF, taken in part from Freitas (1980d), are listed below roughly in order of increasingly sophisticated capability: robot language systems, product assembly, product inspection and quality control, product modification, product repair, product adjustment, product improvement; remedial action by reason of emergency or subtle hazard, robot self-replication. It is assumed in each case that the impediments to meeting these requirements (e.g., control techniques, "packaging" to withstand hostile ambient environments, etc.) will somehow be overcome. The first three functional requirements are described briefly below, followed by a general discussion of the more advanced requirements.
Robot control languages. Numerous machine languages exist for the control of semiautomated machine tools (Lindberg, 1977). These include APT (automatic programming tool) and ICAM (integrated computer aided manufacturing). McDonnell Douglas Aircraft Company has recently extended APT to MCL (manufacturing control language) in order to program a Cincinnati Milacron T3 robot to rivet sheet metal. Higher-level robot control languages, obvious requirements for advanced automated space systems, include VAL (versatile assembly language) for the Puma robot and "HELP" for the Pragmac robot (Donata and Camera, 1980). The problem of extending high-level languages from comparatively simple machine tools to more sophisticated multiaxis integrated robot systems which may be found in future automated space factories must be viewed as a top priority research item.
Product assembly. At SRI International, requirements for the five basic operations in factory assembly have been evaluated by Rosen et al. (1976). These include (1) bin picking, (2) servoing with visual feedback, (3) sensor-controlled manipulation, (4) training aids, and (5) manipulator path control.
The team has recognized the need for improved performance in bin picking of, say,assorted cast basalt and metal objects. Multiple electromagnetic end-effectors certainly could pick out just the metal casings. Variably energized end-effectors might be used to separate and select metal parts of varying magnetic susceptibility randomly arranged in a bin (i.e., aluminum vs iron vs titanium parts). But general bin picking from random parts assortments is not yet possible, though it might be essential in a fully automated SMF operation.
SRI has applied visual servoing by combining a General Electric television (100 X 100 element solid-state) camera with an air-powered bolt driver incorporated into an end effector. Three-dimensional cameras may be required for highly contoured objects fabricated in space (Agin, 1980; Yachida and Tsuji, 1980). Such cameras have already been applied to automated bin selection tasks by the Solid Photography Company in Melville, New York.
Computer-vision technology needs to be merged with discoveries from biological studies. Automatic gain control, gray-scale imaging, and feature detection must be included in computer-vision Automation Anywhere technology if robot autonomy is the goal. Parallel computer-control systems will ensure the speed of reaction and self-preservation "instincts" required for truly autonomous robots, but will require a decrease in existing computer memories both in size and access time by several orders of magnitude. Consideration should be given to associate and parallel memories to couple perceptions to the knowledge base in real time.
To achieve sensor-controlled manipulation, somewhat greater precision Is required of robot arms than can be obtained now. Present-day Unimates (control and precision of 2.5 mm) have been used in a one-sided riveting operation using strain-gauge sensing of the rivet gun mandrel, but there is still a need for more rapid finding, insertion, and fastening by passive accommodation, servo adjustment, and search algorithms. A novel "eye-in-the-hand" adaptation for rapid assembly in space may utilize acoustic sensors. The Polaroid Corporation in 1980 applied its camera ranger to end-effectors for tool proximity sensing. The unit emits a millisecond pulse consisting of four ultrasonic frequencies (50, 53, 57, and 60 kHz). Ultrasonic techniques are potentially quite useful in air or other fluid-filled bays in nonterrestrial manufacturing facilities, especially in view of the acoustic positioning systems developed by the Jet Propulsion Laboratory for containerless melt manipulation. Under vacuum conditions when precise positioning is necessary, laser interferometry may provide the answer (Barlunann, 1980).
Regarding training aids, more sophisticated coordinate transformation programs are required to operate manipulators for diverse tasks. A possibility for the future is "show and tell," a new technique for robot training (see chapter 6). Ultimately, a robot itself could train future-generation machines through some means of "training-by-doing." A related issue - the problem of robot obsolescence - will not be trivial.
Finally, manipulator path control should be fully automated in SMF where, for example, rock melts must be transported along smoothly controlled paths (see the discussion of basalt fiber spinning in section 4.2.2). In the manufacture of bearings or fibers where high-speed trajectories are involved, manipulator halts at corners must be avoided by developing better path control strategies. In the near-term, it may be possible to extend the capabilities of the Unimate:PDP-11/40 couple. For every machine proposed for the SMF, including the starting kit extruder, it is simplest to use a coordinate system based on that machine to interact with robot manipulators continuously to redefine forbidden regions and motions. Thus, a major requirement in robot factory assembly is to specify the coordinate systems of the component machines.
Product inspection and quality control. The need for visual methods of inspection and quality control by automata must be defined for each class of SMF product envisioned. For instance, the application of electroforming on the Moon to produce thin-walled fragile shapes, aluminum ribbon extrusion, or internal milling of Shuttle tanks, definitely demands inspection and quality control. Terrestrial automated inspection systems currently are in use at General Motors, Western Electric, General Electric, Lockheed Recognition Systems, Hitachi Corporation, SRI International, and Auto-Place Corporation. A detailed synthesis of the vision requirements for each is given by Van der Brug and Naget (1979). Off-the-shelf television systems with potential for robotics applications already provide measurements to 1 part in 1000 of the height of the TV image, e.g., the EyeCom Automated Parts Measurement System manufactured by Special Data Systems, Inc. in Goleta, California. Finally, the use of fiber optics in quality control, as demonstrated by Systems now in use by Galileo Electronics, Inc., warrants further development.
Advanced functions and recommendations The needs of space manufacturing for automated product modification repair, adjustment and improvement, as well as robot adaptation to emergencies and self-replication, depend in large part on the capabilities of future automata control system and the environment in which they are applied. The hazards of space to human beings are well known, whereas the impact on robot systems is less well understood. Potential dangers include rapid pressure changes, spillage of corrosive fluids or hot melts due to vessel rupture, radiation effects from solar flares (e.g., embrittlement), anomalous orbital accelerative perturbations producing force-sensor errors, and illumination-intensity variations caused by space platform tumbling or nutation (producing visual observation problems such as shadow effects in fiber optics sensors).
Robotic intelligence must be vastly increased if these devices are largely to supplant human workers in space. This may be accomplished by deploying a versatile intelligent multipurpose robot or by developing a number of specialized, fixed-action-pattern machines. Multipurpose intelligent robots lie well beyond state-of-the-art robotics technology, yet they still are an important ultimate goal. In the interim, sophisticated fixed-action-pattern robots suitable for restricted task scenarios should be developed. The behavior of such robots would be not entirely different from that of many plants and animals endowed with very sophisticated fixed action patterns or instincts.
Before true machine intelligence can be applied to factories in space, the requirements for automated nonterrestrial manufacturing systems must be determined by an evaluation of the state-of-the-art in this field. A complete and updated computerized library containing abstracts of all available robotics research and applications publications, accessible through ARPANET, should be implemented to enhance automation technology transfer. Among the subject categories which should be emphasized are controls, arm/work envelopes, robot adaptability, applications, and costs. Knowledgeability in the field requires contact with finns listed below to better understand how solutions of the practical problems of today can be extrapolated to help solve those of tomorrow: Unimation, Inc.; Cincinnati Milacron; ASEA, Inc.; Prab Conveyors, Inc.; Planet Corporation; Devilbiss/Trallfa; Nordson Corporation; Binks, Inc.; Thermwood Machinery Corporation; Production Automation Corporation; AutoPlace Company; Modular Machine Company; Seiko Instruments, Inc.;Jones Oglaend Corporation; Fujitsu Fanuc Corporation; Okuma Machinery Corporation; Advanced Robotics Corporation; Hitachi Corporation;and Benson-Varian Corporation.