As a generalist, unencumbered by details, I can imagine construction robots similar to those used in factories but each mounted on a second robot which establishes its relative location instead of on the fixed frame used in most production lines. Locater robots may connect to each other mechanically or informationally but typically the construction robots will work in teams or small sets. The alternative to multiple robots is special robots built for each task as they are in existing factories. I think of generalized robots as having three arms, one for sensors, one for manipulation, and one for tools, thus triangular and able to rotate on its locater base. By combining the functional and computing ability of several generalized robots, many special functions can be implemented.

Robots are essential in the space environment for all the usual reasons: it is dangerous, difficult to get to, expensive to make round trips, and filled with the kind of unknowns that make work slow and experimental. Since the goal is construction I want to elaborate on that function. In factories robots specialize in the kind of simple repetitive actions that can be reduced to bnc programs. Construction, although it has repetitive aspects is more of a sequence of actions. To make the thought experiment think of constructing a model airplane. The first step is to establish a workspace sufficient to the finished model. The robot would have to be able to also construct a 3d plan. A person would look at the picture on the box but the robot, lacking human thought would have to build the plan before it could build the plane. A person might just dive in but the robot would have to inventory the parts and internalize the part list. Likewise the instructions provided for a person would not be adequate but a detailed set of instructions would have to be developed. A robot might have a general set of instructions and might be able to compare the written instructions generally for things like sequence and orientation but the human instructions would probably be a near total loss for the robot. The robots plan would have to be a kind of exploded view identifying each part. The robot would have to identify the part, orient it correctly with respect to the workspace and glue it in place. My theoretical robot has three arms which perform these functions. Presuming we input the parts list and the 3d plan the robot team must just hold the work piece, a robot picks up a part, identifies it against the list and the 3d plan, orients itself and the part so that it can be placed, and glues it on. Since a lot of research has been done on visual identification, we might find a usable program for that. Manipulation, insertion, and gluing are things industrial robots do all the time. Getting the robot to build to a programmed plan is at least feasible � getting it to do the work of a 10 yr old is going to be a lot more difficult. It may be that there is a more simple-minded approach that would put the parts list in assembly order and build it up but without some feedback to a plan the potential for bizarre results is probably large. Weather or not our robot ever learns to construct its own plan is moot � all we need to demonstrate is that a robot can build a model airplane. Plans can be made in any place where humans can develop them.

Our proposed robot is close kin to the industrial robots used to assemble, join, paint, and handle parts on an assembly line, except of course space doesn't even have down, much less a line for assembly. First consideration is that robots must communicate not only on a processor level but also physically so they can generate a common grid. Living creatures have advanced sensory feedback, computers by contrast have very limited awareness of their environment. The assembly line acts as a reference for the actions of a robot so that they are absolute in the defined context. No robot of current design is going to have the kind of situational response of a rat or pigeon, but because space is by definition mostly empty we can pre-define an assembly environment.(the �no cats� rule) Probably it will be useful to make each robot in two parts, one for transport and the other for manipulation. The transport part can communicate with others to develop and maintain a grid (orientation). Sometimes it might be useful to connect transports with a rod or beam to create a rigid relationship so our design will include this as a design criteria (perching ie scaffold mounted). Then each will have to be mobile, at least enough for making location adjustments (mobility). The manipulation part of the robot must have arms for manipulation, sensory feedback, and tool use or application. Currently I think three arms, but multi-function arms are possible by using different 'hands'. Swapping 'hands' is problematic because in space we probably want all parts to stay attached. There are two types of arms: articulated, and telescoping. I think maintaining working coordinates at the hand (coordination) may be easier with the telescoping arm and also it may be somewhat stronger than the articulated design. Articulation is useful to humans because we can work on something in front without our arms being in the way but by using a second arm for sensing the robot does not have a front and center perspective, and less feedback may be required to maintain coordination. By using the robots in groups we can make use of multiple hands and triangulation to assist sensory feedback and also eliminate the need to reach behind anything.

Our robot is built with many assumptions and a few more are needed to develop a test case. Before building a space station in orbit we need to be able to construct something in a test environment. I propose a plastic model. The operations are to identify and sort parts, trim or modify parts, manipulate and match parts, glue parts together, and review the assembly process. Identification of parts requires that the robot must build, or have access to, a parts list that includes patterns and part numbers. Visual identification may be a goal but 'touch' may by sufficient. If the part has a sprue or part of a sprue it must be identified and trimmed. Assembly sequence may be supplied although the goal here is to have a general sequence which can be modified to build a particular model. The parts are selected according to the sequence, manipulated to match and then glued together. So the tool arm has a glue applicator, the manipulator is some kind of gripper, and the sensory arm has a whisker to feel the shape of the part. At least two or three robots must work together and there must be a computer model which is compared to the plastic model to assess progress and part identification. The assembly robots do not need to be autonomous, instead they need to network and perform simple assembly tasks. Somewhere in the network there may eventually be more autonomy but for now we are going to perform that function by providing programming. Anyone who has seen robots in an assembly line will accept that this is possible.

The next stage of the process is to process raw materials and manufacture parts. One of the appealing things about space is the variety of materials that can be found. Meteorites may be metallic or glassy and may even contain liquids and gasses. Some of the early proposals involve concentrating sunlight to melt and cast materials on the surface of some body and certainly before we try it is hard to argue in favor of a certain method but since my proposal is for the use of small robots I favor forming small elements by some tool based method, possibly sintering or electroforming. Metallic parts can be welded and non-metallic parts can be joined by the same process that formed them.

The decomposition for robots includes: arms and tools, bodies and supports, carriers, frameworks, programs, operations, maintenance, transport, storage, building, and recycling.