This is my personal framework for AI experiments in Factorio. Everything is subject to change.
Ember autopilot works by binding agent programs to players to perform certain tasks. The autopilot is mainly controlled through a selection item named the Ember Controller.
One critical feature of Ember AutoPilot is it's ability to move autonomously, this is where the main focus of the development lies right now. At this point it has capabilities to do basic pathfinding, but it's still rough around the edges.
You can install this mod through the built-in modloader of Factorio under the name "Ember Autopilot". Alternatively, you can manually install it using git or by downloading the source code in .zip format.
Mod folder (Windows): %appdata%\Factorio\mods
Mod folder (Mac OS X): ~/Library/Application\ Support/factorio/mods
Mod folder (Linux): ~/.factorio/mods
cd <path to factorio mods>
git clone https://github.com/Diordany/factorio-ember-autopilot.git ember-autopilot_0.6.2
Just save the .zip file to the mod directory of Factorio as ember-autopilot_0.6.2.zip (or leave the name as is if downloading from the releases page or the mod portal).
To get access to the Ember Controller, you'll first need an empty inventory slot, then you can press the launcher button on the top left of your HUD:
Alternatively, you can run the following command:
/ember-controller
To get rid of the controller, just drop it on the ground (belts don't work).
To move your agent, first equip the Ember Controller, then click on your target tile:
The movement mode can be set through the mod settings (per player):
| Option | Description |
|---|---|
| Path BFS | Uses Breadth First Search to generate a path. |
| Walk Straight | Makes the agent walk in straight lines. |
| Factorio Pathfinder | Uses the pathfinder provided by Factorio. |
| Wander | Makes the agent wander aimlessly. |
| Path DFS | Uses Depth First Search to generate a path. |
If you want the agent to walk directly towards the goal position, you're best off using the Walk Straight option. For path generation, the Path BFS option is recommended over Factorio Pathfinder and Path DFS as the latter two are very unreliable.
The moveset setting determines what moveset is used for the search strategies of Path BFS and Path DFS. The names are based on the moveset of chess pieces.
| Option | Description |
|---|---|
| Rook | Includes: north, east, south and west. |
| Queen | Includes: north, northeast, east, southeast, south, southwest, west and northwest. |
| Bishop | Includes: northeast, southeast, southwest and northwest. |
In general, the Rook option is the most useful for the Path BFS and Path DFS modes, since they perform a blind search and don't take the movement cost into account.
To cancel the agent's movement press SHIFT + LMB with the Ember Controller equiped.
Alternatively, you can cancel the agent with:
/ember-stop
When the agent uses the BFS strategy, it tries to find the path with the least amount of actions [1]. This is not necessarily the optimal path [1], since diagonal movement actions cover more distance and thus take longer. Search strategies that address this issue will be part of the next update.
Here's a screenshot of BFS pathfinding with the Queen moveset:
And another one with the Rook moveset:
IMPORTANT: keep in mind that the agent will keep on searching indefinitely if no valid path exists. In that case, stop the agent with SHIFT + LMB.
The built-in pathfinder of Factorio doesn't seem to be particularly useful for this application. The paths are rarely optimal and route straight through collidable entities (even if the collision mask is given) as seen here:
With the DFS strategy, the agent follows branches as far as it possible can, which is a big problem for vast or infinite search spaces [1] (like the Factorio world):
Avoid this option unless you're in a closed environment. A bounded version is probably going to be introduced at some point.
The amount of search nodes processed per tick can be set to tweak the search performance.
Please see the contibution guidelines if you want to work on Ember Autopilot.
[1] S. Russel and P. Norvig, "Solving Problems by Searching", in Artificial Intelligence A Modern Approach, 3rd ed. Harlow: Pearson Education Limited, 2016, ch. 3, pp. 64-119.