RUBE is a tribute to Rube Goldberg, creator of all those unique cartoon contrivances, who showed the world that anything can be done in a more complicated fashion than necessary.
The RUBE language is a "bully" cellular automaton: certain state transitions force other, non-adjacent cells to assume certain state transitions. Although this may sound like an interesting notion, the number of interactions climbs quickly as more objects are added, and things get much messier than they normally would in a regular cellular automaton, like John Conway's Game of Life.
Also unlike a real cellular automaton, RUBE supports rudimentary output functionality; input was planned, but has not yet been implemented.
Because implementing an algorithm in RUBE is a clumsy job of co-ordinating concurrently operating sections of the playfield, few non-trivial RUBE programs exist. In particular, it has never to the author's knowledge been shown that RUBE is Turing-complete. On the other hand, it has never been shown that it's not.
See rube.txt for the original description of the RUBE interpreter. The only point that perhaps needs clarification is that the limitation of the playfield to 80x25 is a limitation of the implementation, not of the language. The language imposes no bounds on the size of a RUBE playfield.