Page:Advanced Automation for Space Missions.djvu/274



According to the baseline mission for a growing, self-replicating Lunar Manufacturing Facility (LMF) presented in section 5.3.4, a 100-ton seed is dropped to the lunar surface and thereupon unpacks itself, sets up the initial factory complex, and then proceeds to produce more of itself (or any other desired output). Clearly, the level of automation and machine intelligence required lies beyond current state-of-the-art, though not beyond the projected state-of-the-art two or three decades hence. Because of the already challenging design problem, it is highly desirable to keep all seed systems as simple as possible in both structure and function. This should help reduce the risk of partial or total system failure and make closure less difficult to achieve at all levels.

One of the more complicated pieces of hardware from the Al standpoint is the "camera eyes" and pattern recognition routines (visual sensing) that may be needed. Although it is possible that standardized robot camera eyes may be developed, it is more likely that each particular application will demand its own unique set of requirements, thus greatly reducing or eliminating any gains in simplicity of camera design. The pragmatic industrial approach (Kincaid et al., 1980) and design philosophy in these cases, especially in the area of computer vision, is to: (1) simplify, (2) use unconventional solutions, and (3) "cheat (i.e., solve another problem). It may be that the best way to handle the problem of computer vision is to find a way to largely avoid it altogether.

When the seed unpacks itself it opens into a rather wild environment full of hills, bumps, ledges, crevasses, boulders, craters, and rocks. Surface navigation by mobile robots will be a serious challenge to Al technology. How will a machine know where it is, what the terrain ahead may be like, or how to get home? Laser tracking is one possibility, but probably too complicated when out of line of sight. Pattern recognition of geological and geographical landmarks is another possibility, but there are at least three serious deficiencies associated with this solution. First, the pattern recognition routines must be extremely sophisticated and the sensor very high in resolution and in the ranges of illumination that may be accommodated. Second, to recall how to get home after a lengthy perambulation across the lunar surface may require vast amounts of onboard computer memory. Every turn, every detour, every move the robot makes must be recorded, analyzed for spatial displacement geometry, and the present-position pointer augmented against the stored features maps and correlated with the geographic images received through the vision sensors to plot the shortest route home to avoid the inefficiency of retracing the original physical path. Third, since exploration, development, and construction operations are always in progress around the site, each robot would need a memory capacity sufficient to recall in detail all changes in the landscape between the last series of explorations and the present one - the view is always changing. It may not be practical to design this much Al into each mobile robot, nor to require the central computer to exercise full teleoperator control of a large fleet of nonautonomous mobile robots.

5B.1 The Transponder Network

One way to achieve accurate positioning of all mobile robots while retaining their navigational autonomy is to employ a transponder system operating in the gigahertz frequency range. Much like the LORAN and NAVSTAR systems on Earth, these radar beacons would permit the accurate determination of position by simple triangulation for mobile robot devices located anywhere in the vicinity of the seed. A frequency of perhaps 30 GHz, easily within the range of current technology, would be required for 1-cm positioning accuracy. The transponder system could be orbital-based, but for the present design a ground-based system has been assumed with at most a single satellite for purposes of initial calibration.

When the seed unpacks, its first task is to unfurl the "home base" transponder. Power consumption has not been examined in detail but should not exceed 100 W, the amount supplied by a 1 m2 solar panel. The next step is to establish an accurate navigational baseline between the home transponder and a reference transponder some distance away, perhaps using a relatively simple nonlaser surveyor's transit. A second baseline is similarly established in some other direction, and the whole system then calibrated and synchronized to coherence. Thus deployed, a local radio navigation grid exists which can fix the position of any appropriately equipped receiver to within 1-cm accuracy, horizontally or vertically, anywhere near the seed.

Since the transponder operates on line-of-sight, each transmitter must be placed a certain distance above the ground in order to "see" the entire area for which it is responsible. The general horizon distance formula is