Wilson’s algorithm uses loop-erased random walks to generate a uniform spanning tree — an unbiased sample of all possible spanning trees. Most other maze generation algorithms, such as Prim’s, random traversal and randomized depth-first traversal, do not have this beautiful property.
The algorithm initializes the maze with an arbitrary starting cell. Then, a new cell is added to the maze, initiating a random walk (shown in magenta). The random walk continues until it reconnects with the existing maze (shown in white). However, if the random walk intersects itself, the resulting loop is erased before the random walk continues.
Initially, the algorithm can be frustratingly slow to watch, as the early random walks are unlikely to reconnect with the small existing maze. However, as the maze grows, the random walks become more likely to collide with the maze and the algorithm accelerates dramatically.
The global structure of the maze can be more easily seen by flooding it with color.