Explaining the Entombed Algorithm

Abstract: In \cite{entombed}, John Aycock and Tara Copplestone pose an open question, namely the explanation of the mysterious lookup table used in the Entombed Game's Algorithm for two dimensional maze generation. The question attracted media attention (BBC etc) and was open until today. This paper answers this question, explains the algorithm and even extends it to three dimensions.

• The paper clarifies how the lookup table, guided by three invariants, underpins the maze generation in the Atari game Entombed.
• It details a stepwise approach using local neighborhood configurations to decide wall or path placement while preserving maze connectivity.
• The study extends the algorithm to three dimensions, showing that the invariant-based method scales from 32 to 1024 possible configurations.

Analysis of the Entombed Algorithm

This paper, authored by Leon Mächler and David Naccache, provides an in-depth dissection and explanation of the algorithm used in the 1982 Atari game "Entombed" for generating two-dimensional mazes. The algorithm has intrigued both researchers and the public due to its enigmatic reliance on a predefined lookup table and local information to generate complex, playable mazes. This study resolves the open question posed by Aycock and Copplestone regarding the origin and function of the lookup table within the Entombed algorithm and further extends the algorithm to three dimensions.

Key Components of the Entombed Algorithm

The Entombed algorithm operates by sequentially constructing lines of mazes, block by block, using a decision-making process based solely on the configuration of five neighboring blocks. Each block is either a wall or a path, represented by binary values. The algorithm utilizes a lookup table for determining whether a wall or path should be placed at each position, contingent upon one of 32 possible configurations of neighboring blocks. Notable within this process is the use of randomness where multiple choices maintain certain invariants. These invariants prevent the formation of trivial patterns and ensure connectivity throughout the maze.

Explanation of the Lookup Table

The central innovation of this research is elucidating the structurally elaborate lookup table, which is based on maintaining three logical invariants:

1. Invariant 1: No $2 \times 2$ square should comprise identical elements (all walls or all paths).
2. Invariant 2: No wall or path can start or end with a thickness of one.
3. Invariant 3: Paths in one line must connect to a path in the following line.

Each decision within the lookup table aligns with these invariants, ensuring efficient maze generation that preserves structural integrity and connectivity.

Crucially, the researchers identify that the entirety of the lookup table can be accounted for by these three invariants, with the exception of a negligible context where the invariants appear in conflict. The paper illustrates how such conflicts can be resolved through additional rules, ensuring the invariants are upheld even under seemingly contradictory circumstances.

Extension to Three Dimensions

The authors successfully extend the algorithm to generate three-dimensional mazes by maintaining the same set of invariants. This extension necessitates a more complex neighboring context due to additional dimensional factors, resulting in an expansion of possible decisions from 32 to 1024.

In extending to three dimensions, the paper explains additional conflicts that occur but notes that a similar ratio of rules to conflicts as found in the two-dimensional case still largely applies. Future iterations of the algorithm may modify the size of the context or adjust the randomness to reduce invariant violations in multi-dimensional contexts.

Implications and Speculation on Future Research

The paper exemplifies the utilization of minimal local information to produce complex outputs, indicative of approaches that could influence procedural generation in constrained environments. As computational methods in AI advance, the precision in algorithmic efficiency demonstrated by the Entombed algorithm might find applicability in broader fields such as automated planning, simulation, and even neural network design. The exploration into higher-dimensional extensions opens the door to complex spatiotemporal problem-solving frameworks, where dimensional scalability is paramount. Such methodologies could contribute to AI advancements in autonomous navigation, complex system modeling, and virtual reality environments.

Future research may explore the optimization of context-based lookup tables across various applications, exploring avenues for automatic generation of logic tables based on desired invariants within procedural systems.