« PreviousContinue »
The beneficial impact of such a step for the industrial economy, for other branches of government, for the public well-being, and for NASA's own future effectiveness in an era of tight budgets is likely to be substantial.
contact with and validation of advanced machine intelligence techniques. Flight minicomputer memories are typically at 16,000 or 21,000 words, enormously restricting options. (For example, a very large number of scientific targets on Jupiter and the Galilean satellites, which otherwise could be acquired, had to be abandoned because of the memory limitations of the Voyager onboard computer.) But million byte memories are now routinely employed and, once space-qualified, could provide enormous flexibility.
We, the NASA Study Group, here state our overall conclusions and recommendations. Our report is complete with supporting documentation leading to these conclusions and recommendations.
Conclusion 1. NASA is 5 to 15 years behind the leading edge in computer science and technology.
Because of the long lead times in the planning cycle, many decisions relating to computers are made five to seven years before launch. Often, the computer technology involved is badly obsolete at the time hardware is frozen. Further, no deliberate effort is made to provide flexibility for software developments in the long time interval before mission operations. (Uplinking mission programs after launch is a small but significant step in the right direction.)
Conclusion 3. The overall importance of machine intelligence and robotics for NASA has not been widely appreciated within the agency, and NASA has made no serious effort to attract bright, young scientists in these fields.
There are some examples of excellence, but in general we find NASA's use of computer technology disappointing. NASA installations still employ punched-card-based batch processing and obsolete machine languages. There is no NASA nationwide computer network and no widespread time-sharing use of computers. Although Viking was a brilliant technological success, given its design limitations, Viking's use of robotics technology and in situ programming was rudimentary. These techniques must be greatly advanced for the complex missions of the future, both planetary and Earth orbital. Most Earth-satellite and much planetary exploration imaging data remains unanalyzed because of the absence of automated systems capable of performing content analyses. Even missions being planned for the 1980s are being designed almost exclusively for traditional data collection with little apparent provision being made for automated extraction of content information.
In 1978/1979, the Space Systems and Technology Advisory Committee of the NASA Advisory Council had 40 members. Not one was a computer scientist, although two had peripherally related interests. Few, if any, of the best computer science PhDs from the leading academic institutions in the field work for NASA. There is a looped causality with NASA's general backwardness in computer science (Conclusion 1): An improvement of the quality of computer science at NASA cannot be accomplished without high quality professionals; but such professionals cannot be attracted without up-to-date facilities and the mandate to work at the leading edge of the field.
Conclusion 2. Technology decisions are, to much too great a degree, dictated by specific mission goals, powerfully impeding NASA utilization of modern computer science and technology. Unlike its pioneering work in other areas of science and technology, NASA's use of computer science and machine intelligence has been conservative and unimaginative.
The problems summarized in Conclusions 1 and 3 cannot be solved separately.
Conclusion 4. The advances and developments in machine intelligence and robotics needed to make future space missions economical and feasible will not happen without a major longterm commitment and centralized, coordinated support.
Strict funding limitations and an understandable aversion to mission failure cause mission directors to settle for proven but obsolete and, ironically, often very expensive technologies and systems. As machine intelligence and robotics continue to advance outside of NASA, the consequences of these traditions for higher cost and less efficient data return and analysis become more glaring. The inertial fixation on 15-year-old technologies, including slow processors and very limited memories, strongly inhibit NASA
A table of various planned future space missions and an estimate of technology development efforts needed to automate their system functions was given in Section IV (see Table 4-1). Without these automatic system functions, many of the missions will not be economically and/or technologically feasible.
missions, and should emphasize research programs with a multimission focus.
Recommendation 1. NASA should adopt a policy of vigorous and imaginative research in computer science, machine intelligence, and robotics in support of broad NASA objectives.
A balance is needed onboard NASA spacecraft between distributed microprocessors and a centralized computer. Although function-directed distribution of processors might be useful, such architectures should not preclude the use of these computing resources for unanticipated needs. Distributed computer concepts emphasizing "fail-safe” performance should receive increased attention. For example, in the case of failure of a computer chip or a unit, a long-term goal is to effect migration of the program and data to other working parts of the systems. Such fail-safe systems require innovative architectures yet to be developed. Dynamically reconfigurable processors with large redundancy are badly needed in NASA.
The problems summarized in the preceding list of conclusions have solutions. They require, most of all, an awareness that the problems exist and a commitment of resources to solve them. Table 6-1 gives the published R&D budgets of the seven largest computer corporations in the United States. In all cases, the total R&D spending is greater than 42% of total profits. The advanced R&D budget would be only a fraction of this amount. Leading corporations in computer science and technology characteristically spend 5.4 percent of gross earnings on relevant research and development. The same percentage of NASA's annual expenditure in computer-related activities would suggest an annual NASA budget for research in computer science, machine intelligence, and robotics approaching one hundred million dollars. An expenditure of half that would equal the combined annual budget for this field for ARPA and the National Science Foundation. If NASA were selected as lead agency (or lead civilian agency) for federal research and development in computer science and technology, such amounts might not be at all impractical. Any significant expenditures should have detectable benefits in three to five years, and very dramatic improvements in NASA programs in 10 years. If NASA were to play such a lead agency role, one of its responsibilities would be to study the long-term implications for individuals and for society of major advances in machine intelligence and robotics.
NASA relies on 256-bit computer memory chips; 16,000 bit and 64,000 bit chips are currently available. A millionbit chip is expected to be available within a few years. The cost of space-qualification of computer hardware may be very high, but the possibility exists that high informationdensity chips may already work acceptably in the space environment. We recommend that NASA perform space qualification tests on the Shuttle of multiple batches of existing microprocessors and memory chips.
These two examples of developments in computer science and technology will have applications to many NASA missions. We also recommend a transitional period in spacecraft computer system design in which existing miniprocessors and new microprocessors are both utilized, the former as a conservative guarantor of reliability, the latter as an aperture to the future.
Recommendation 2. NASA should introduce advanced computer science technology to its Earth orbital and planetary
Table 6-1. R&D of the Big Seven Computer Companies
168 152 118 122 73 80
157 134 144 215
a program in computer sciences, NASA can ensure that there is a rapid transfer of new technology to NASA programs. Space exploration offers a unique environment in which to develop and test advanced concepts in this discipline.
In planetary exploration, “... it is clear ... that more advanced mission techniques and instrumentation are required to fulfill the science strategy and achieve the objectives..." of intensive study of a planet.8 Surface rovers and return-sample missions will be required to meet the science goals for Mars, the Galilean satellites of Jupiter, Titan, and perhaps Venus, as well as for investigation of such specific locations on the lunar surface as putative volatile-rich deposits at permanently shaded regions of the poles. With the exception of the Lunakhod and other Luna-class missions of the Soviet Union, there is little experience with such systems. Because of the long lead times and the complex nature of rover missions, they provide an ideal testing ground for the implementation of the multimission focus of some of our recommendations.
Recommendation 3. Mission objectives should be designed flexibly to take advantage of existing and likely future technological opportunities.
This leads to the following specific recommendation: NASA should consider Computer Science and Technology sufficiently vital to its goals to treat the subject as an independent area of study. The specific concerns of this field, enumerated below, should become research and technology issues within NASA on the same basis as propulsion technology, materials science, planetary science, atmospheric physics, etc. This means the creation of a discipline office for computer science with interests in the major subdisciplines of the field and with appropriate contacts within NASA. A suitable budget and program of research and technology grants and contracts would provide the focus in this field the Study Group has found lacking in NASA. On the one hand, it would help make the outstanding workers in the field aware of and interested in serving NASA's needs. Graduate students participating in such a research program would become a source of future employees for NASA centers and contractors. On the other hand, it would provide NASA Headquarters with a better awareness of the potential contributions of computer science to its programs. To be effective, the initial operating budget of such a program should not be below 10 million dollars a year, with a long-term commitment for at least a constant level of funding in real dollars.
Hardware should be designed to exploit state-of-the-art software and likely near future software developments. Adoption of this recommendation implies a careful reexamination of missions currently in the planning stages. This recommendation applies not only to spacecraft systems but to ground-based computer systems as well. The man/ machine interface, both in Shuttle systems and in mission operations ground equipment, has not, in our opinion, been optimized. In routine mission operations, particularly in mission crisis management, there is a severe short-term competition for human attention and intellectual resources. The problem is a combinatorial one, requiring systematic and exhaustive failure-mode analysis, which can be optimally provided by computer systems, via a probability analysis, analogous to existing computer programs in medical diagnosis. In addition to their value in crisis management, such computer systems will lead to the optimization of subsequent missions.
Most of the fundamental research under such a program would be carried out at universities and at appropriate NASA centers. Collaboration with industry should be encouraged to expedite technology transfer. To meet the emerging mission requirements, parallel advanced development programs within all of NASA's mission offices are required.
Recommendation 4. NASA should adopt the following plan of action:
(a) Establish a focus for computer science and technology
at NASA Headquarters for coordinating R&D activities.
Following is a list of problem areas that should set some goals for both the basic science research program and the advanced development effort:
The pace of advance in computer science and technology is so great that even experts in the field have difficulty keeping up with advances and fully utilizing them. The problem is, of course, much more severe for those who are not experts in the field. By establishing
Smart sensing; automated content analysis; stereo mapping for eventual Earth and planetary applications.
• Manipulator design, particularly for autonomous
use, including structures and effectors, force and touch detectors.
8 Ibid, p. 39.
(d) NASA should form a task group to examine the
desirability, feasibility, and general specification of an all-digital, text-handling, intelligent communication system.
The current university-based support from NSF and ARPA in computer science and machine intelligence is about 15 million dollars each annually. The level of university funding recommended here would be larger by about 30 percent, allowing NASA to compete effectively for the best talent and ideas. Parallel programs conducted by NASA program offices, which would be based strongly at NASA centers and industry, would approximately double the support requirement. The total support might eventually approach the 100 million dollar level, if NASA were seriously to pursue a broad program of research in computer science.
A significant amount of NASA's budget is spent in the transfer of information among a very complex, geographically, and institutionally disparate set of groups that need to exchange messages, ideas, requirements, and documents quickly to keep informed, to plan activities, and to arrive at decisions.
(b) Augment the advisory structure of NASA by adding
computer scientists to implement the foregoing recommendations.
Based on a rough estimate, we predict that such an all-digital network would lead to significant improvements over the present method of carrying out these functions. In addition to the cost savings, there would be improvements in performance. Although it would not eliminate the use of paper and meetings as a means of communication, it would save tons of paper and millions of man-miles of energy-consuming travel. This system would facilitate and improve the participation of scientists in all phases of missions as well as enhance their ability to extract the most value from postmission data analysis.
NASA is far enough behind the leading edge of the computer science field that major improvements in its operations can be made immediately using existing
making; and a model for other public and private organizations, scientific, technological, and industrial.
The implementation of such a system would not be predicated on new developments in artificial intelligence, but on the tools that are in common use at artificial intelligence nodes of the ARPA network and are part of the developing technology of digital information and word processing. If such a development were carried out, it would provide the data base for sophisticated techniques, as they become available, for information retrieval, semantic search, and decision
The task group to investigate this development should include elements of NASA management, mission planning and operations, scientific investigators, and information scientists, as well as specialists in artificial intelligence.