|
|
|
|
Toshihiro Matsui, Electrotechnical Laboratory
Suppose you have a bunch of mechanical parts of a particular assembly, say, a model aircraft engine, but you don't have the plan or any description about how to assemble them together. This is often the case when you disassemble equipments around you. Besides that, if you are a machine designer, you must be very interested in the possiblity of the parts you are designing to be conflicting with other parts.
This Assembly Goal State Planner tries to put the parts together to find the most plausible assembly. In the model aircraft engine example, it finds all fitting pairs between protruding cylinder and concavity with the same diameter. For the links from the piston-pin, conrod, crank, to the bearing, symbolic mathematical formulae processing is applied to ananyze the propagation of constraint caused by the degrees of freedom left undetermined in a single pair of links. For non-cylindrical objects, it tries to maximize the contacting area or to minimize the volume of their convex-hull. These strategies are natural heuristics since we use the same techinique when we play with block structured objects, such as lego blocks.
We cannot expect the planner to always discover the correct solution, but we can utilize it for finding interference between parts, and to see the final image from parts to check the validity of each part in the assembly.
In the following picture, we can see most of the parts are correctly located and oriented, but the orientation of the carburetor could not be determined. Stronger constraint is needed to determine it.
This planner was built while the author was at the year of leave
to the Stanford Computer Science Robotics Laboratory, directored by
Professor Jean-Claude Latombe, in 1991 and 1992.
