smartive / rust-sokoban-level-generator

A procedural level generator for "sokoban".

Sokoban Level Generator in Rust

This is a sokoban level generator algorithm written in rust to procedurally generate levels for the game Sokoban.

The algorithm is based on the following resources:

• Procedural Generation of Sokoban Levels; a technical report by Joshua Taylor and Ian Parberry.
• Sokoban-Level-Generator; a javascript implementation by AlliBalliBaba.

Usage

In order to use the generator, add this library (via git) to your project with cargo. Then, import the level generator and use it to generate a level. The level_to_string method allows you to print the level in the JSoko Level Format.

use sokoban_level_generator::{generate_level, level_to_string};

fn main() {
    let level = generate_level(4, 4, 3);
    println!("Print level");
    println!("{}", level_to_string(&level));
}

The algorithm uses randomness to generate levels and needs to create the "farthest state possible" for boxes. As such, generation with more boxes can take a very long time. See the benchmarks section for more information.

Warning: Currently, there is no abort limit. So, trying to generate impossible levels (e.g. 1x1 with 4 boxes) will result in an infinite loop.

Algorithm

The algorithm performs the following (high-level) steps:

1. Generate an empty level with width * 3 and height * 3 dimensions, since every "room" is 3x3.
2. Generate rooms in the level (by using the described templates in the paper); rooms are randomly selected and rotated
3. After the level is generated, check the level for:
   • Enough space
   • Enough goal possibilities (to place goals)
   • No large spaces (this does not generate interesting levels)
   • All floors are connected
   • No surrounded floors
4. Generate random goal locations
5. Calculate the "farthest state possible" for each box/player position
6. Return the farthest possible level

Benchmarks

Benchmarks show that the algorithm is very fast for small levels, but can take a very long time for larger levels. The biggest impact on generation time is the number of boxes.

The result in short: feasible, interesting levels can be generated with up to 4 boxes in a reasonable amount of time. The height / width of the level should be kept below or equal to 3. If 3 or fewer boxes are used, the level can be up to 4x4 in size. Bigger sizes must be experimented with.

The following tables shows the generation time for different level sizes.

1 Box

Width	Height	Median Time
1	1	5.7706 µs
1	2	19.403 µs
2	2	84.277 µs
2	3	4.3456 ms
3	3	1.3252 ms
3	4	6.3146 ms
4	4	33.759 ms

2 Boxes

Width	Height	Median Time
1	2	36.320 µs
2	2	241.53 ms
3	2	798.81 µs
3	3	2.9076 ms
4	3	10.201 ms
4	4	42.901 ms

3 Boxes

Width	Height	Median Time
2	3	4.3456 ms
3	3	20.324 ms
3	4	50.916 ms
4	4	227.51 ms

Violin Plot (1-3 Boxes)

As we can see in the violin plot, the number of boxes has a huge impact on the generation time.