Rog-O-Matic XIV
The program that plays Rogue by itself, developed at Carnegie Mellon University by Michael Mauldin, Guy Jacobson, Andrew Appel and Leonard Hamey. Rog-O-Matic was for a time the strongest known Rogue player, human or machine, and it is one of the earliest substantial pieces of game-playing AI to survive in source form.
Rog-O-Matic drives a real copy of Rogue through a terminal, reading the screen as a player would. The CMU distribution was tied to particular versions of the game: Rogue 3.6 and Rogue 5.2, whose binary came packaged with supporting files for 5.3 (see Rogue 5.2).
The paper
Michael L. Mauldin, Guy Jacobson, Andrew Appel and Leonard Hamey, ROG-O-MATIC: A Belligerent Expert System. Fifth Biennial Conference of the Canadian Society for Computational Studies of Intelligence, London, Ontario, 16 May 1984. An earlier version is Carnegie-Mellon University Technical Report CMU-CS-83-144, July 1983.
The paper describes Rog-O-Matic as an expert system that combines algorithmic knowledge sources with production rules to explore a hostile, partly hidden world: the Dungeons of Doom. It covers the program's architecture, the experts that decide when to fight, flee, explore or use an object, how it learns about monsters from game to game, and how it compares with human players. In games against Rogue 5.2 at CMU from January 1 to February 21, 1983, it had the best median score of any of the strong players it was measured against.
- ROG-O-MATIC: A Belligerent Expert System, Fifth Biennial Conference of the Canadian Society for Computational Studies of Intelligence, 1984.
- ROG-O-MATIC: A Belligerent Expert System, Technical Report CMU-CS-83-144, Carnegie-Mellon University, 1983.
A. K. Dewdney wrote about it in Scientific American's Computer Recreations column, February 1985: “An expert system outperforms mere mortals as it conquers the feared Dungeons of Doom”.
The 2026 restoration
The master branch gathers the work that kept Rog-O-Matic playing after
CMU: Peter Schauer's fixes from 2008, Anthony Molinaro's support for
Rogue 5.4 and his release of it under the GNU GPL, and the project's own
2008 port to Linux, ncurses and 64-bit systems.
In November 2026 it gained rgmloop, which plays many seeded games
unattended, on a pseudo terminal, with memory-checked builds and a record
of where any game that hangs went wrong; rgmrerun, which types a logged
game back into Rogue; and a run of bug fixes found that way.
Repository
https://github.com/rogueforge/rogomatic14. master, the default
branch, carries the current work. The mlm, ant, leland, molinaro
and schauer branches keep the lines of work it was gathered from, and
oldmaster the earlier restoration.
Release history
September 5, 2008
Continuing to patch the source. It now plays the repository versions of Rogue 3.6 and 5.2, and Rogue 5.4 less well; that needs a fix to the message parsing.
August 25, 2008 — a home at RogueForge
After applying Peter Schauer's patches to Rogue 5.2, I tried his patches for Rog-O-Matic XIV too, and watched it become a total winner six times over as it played 10,000 games in about four hours. If nothing else it is a useful tool for testing Rogue, so it now has a home here.