Sturtevant, N. R. (2012). Benchmarks for grid-based pathfinding. IEEE Transactions on Computational Intelligence and AI in Games, 4(2), 144-148. DOI: 10.1109/TCIAIG.2012.2197681
The data is being made available to help improve the consistency of experiments when it comes to gridworlds.
BioWare deserves strong recognition for giving permission for their maps to be distributed for research purposes. We encourage other game companies to follow BioWare's lead and make map test sets available for research and testing purposes.
File Formats
Map FormatAll maps begin with the lines:
type octile
height x
width y
map
where x and y are the repsective height and width of the map.
The map data is stored as an ASCII grid. The following characters are possible:
. - passable terrain
G - passable terrain
@ - out of bounds
O - out of bounds
T - trees (unpassable)
S - swamp (passable from regular terrain)
W - water (traversable, but not passable from terrain)
Scenario Format
Code to load the scenario files is available as part of HOG2, or can be accessed here: ScenarioLoader.cpp and ScenarioLoader.h. While HOG2 is available as LPGL, these two files are offered without any license restrictions.
The scenario files have the following format.
- They begin with the text "version x.x". This document describes version 1.0. The trailing 0 is optional.
- Each line of a scenario has 9 fields:
- The optimal path length is assuming sqrt(2) diagonal costs.
- The optimal path length assumes agents cannot cut corners through walls
- If the map height/width do not match the file, it should be scaled to that size
- (0, 0) is in the upper left corner of the maps
- Technically a single scenario file can have problems from many different maps, but currently every scenario only contains problems from a single map
Bucket | Map | Map width | Map height | Start x-coordinate | Start y-coordinate | Goal x-coordinate | Goal y-coordinate | Optimal length |
0 | maps/dao/arena.map | 49 | 49 | 1 | 11 | 1 | 12 | 1 |
0 | maps/dao/arena.map | 49 | 49 | 1 | 13 | 4 | 12 | 3.41421 |
-
Maze Maps & Benchmark Problems
Nathan R. Sturtevant
Contains 60 maps of size 512x512 and benchmark problem sets. These maps are algorithm-generated mazes with corridor widths of 1, 2, 4, 8, 16, or 32. There are 10 maps and problem sets for each corridor size.
-
Random Obstacle Maps & Benchmark Problems
Nathan R. Sturtevant
Contains 70 maps of size 512x512 and benchmark problem sets. These maps are algorithm-generated by blocking grid cells. Maps contain 10%, 15%, 20%, 25%, 30%, 35%, or 40% blocked cells. There are 10 maps and problem sets for each percentage.
-
Room Maps & Benchmark Problems
Nathan R. Sturtevant
Contains 40 maps of size 512x512 and problem sets. Maps are divided into rooms of size 8x8, 16x16, 32x32, and 64x64. There are 10 maps and problem sets for each room size. Maps with differing room sizes are not scaled: thickness of walls and passages differs.
-
Dragon Age: Origins - Maps & Benchmark Problems
Nathan R. Sturtevant and BioWare Corp
Maps extracted from Dragon Age: Origins with help and explicit permission from BioWare Corp. for use and distribution as benchmark problems.
Contains 156 maps and benchmark problem sets.
-
Warcraft III - Maps & Benchmark Problems
Nathan R. Sturtevant and Blizzard Corp.
Maps extracted from Warcraft III from Blizzard Corp. for use and distribution as benchmark problems.
Contains 36 maps and benchmark problem sets, scaled to 512x512 and converted to a simple grid-based format.
-
Baldur's Gate II - Maps & Benchmark Problems
Nathan R. Sturtevant and BioWare Corp
Maps extracted by Yngvi Björnsson from Baldur's Gate II with explicit permission from BioWare Corp. for use and distribution as benchmark problems.
Contains 75 maps and benchmark problem sets scaled to 512 x 512 and 120 original scale maps.
-
Starcraft - Maps & Benchmark Problems
Nathan R. Sturtevant and Blizzard Corp.
Maps extracted from Starcraft from Blizzard Corp. for use and distribution as benchmark problems.
Contains 75 maps and benchmark problem sets, converted to standard format by Dave Churchill and post-processed to remove all but the largest connected component.