Archipelago - Computer Opponents - Contract Delivery Decisions

Back by popular demand!

Introduction

It's three months since I last made a contribution to Archipelago. I've had some distractions from a couple of other projects but I'll also to admit to getting a little stuck on the computer player automation and decision-making. With the developments in this contribution I think I'm now over that hurdle. Maybe I'll even have a one player version out in time for Christmas!

If you haven't come across Archipelago before, it's a seafaring and trading turn-based strategy game that I'm developing as an open-source project. Players guide their ships around the islands, searching for goods, building ships, trading with the Kingdom settlements and avoiding being chased down by pirates. I've got most of the game mechanics in place and I'm now working on the automation of computer opponents for the one player version.

In previous developments I had given computer opponents the ability to choose between different island search moves, that is, which island to visit and search for resources, or whether to travel and pick up resources found but unclaimed by another player. The avoidance of pirates was also built into this system.

In this contribution I move on to contract delivery decision-making and team-based strategies.

---

Repository

The repository for Archipelago can be found here. Background details of the project and the road-map can be found in the read-me file.
https://github.com/miniature-tiger/archipelago

For a full view of the current state of the game see my github-hosted page. Please note that the game is not yet complete - you can see current progress and test functionality but you will have to wait a little longer to play!
https://miniature-tiger.github.io/archipelago/

---

New Features covered by this Contribution

Contract statistics and ranking of contract moves

Fulfilling contracts for goods issued by the four Kingdom islands is the key method for winning the game. A player must collect goods produced by their resource piece ("resource collection") and deliver an initial amount to the Kingdom island that issued the contract ("resource delivery"). Points are awarded for the length of the trade route and being the first player to contract with an island.

The automated decision making process that I previously developed for island search moves is based upon a probabilty-weighted rating system for each potential move choice, with the score from each outcome multiplied by the probabilty of that outcome arising. The amount of moves that an outcome requires is then used to produce a rating-per-move which allows each different move to be ranked.

The avoidance of pirates was built in to this system by deductions from the ratings based on the probabilty of being caught and defeated in a sea battle and the number of moves required to repair the ship.

This new feature collates statistics on all fulfillable contracts for a player and scores each potential "resource collection" move based on points available from the contract delivery and the number of moves required to complete the delivery. The aim is to have comparable scores to those from island search moves to allow the automation to choose between different move types.

This is most easily explained by an example:

The above screenshot shows the unsorted points-per-move scores for all possible contract-related "resource collection" moves for a single ship on a player turn.

The top three moves that target different resources are then fed into the overall decision making process. More on this below.

Team-based decision-making

At this point I have a method for ranking "island search" moves and a method for ranking "contract resource collection" moves for each ship. The main stumbling block was how to use these rankings to determine which moves an automated player should make, making sure that each of the (possible) three ships pursue different actions and don't all head for the best target.

The solution was to implement a decision-making routine at the start of the turn which took the top three moves of each type for each ship prior to any moves being made. The team-based decision making then chooses three moves, one for each ship, ensuring no overlap between move targets. This allows a team-based strategy to be applied to moves.

There are various methods to implement this strategy, for example maximising the overall score rating across the team or ensuring the best overall move is prioritised. In the first implementation I have used the latter approach. However I plan to use various implementations to give the computer players different strategies and personalities.

This process is repeated after each ship is moved within a turn to incoporate any new information revealed by that move. The prior moves are fixed within the decision making choices to ensure later moves do not repeat earlier moves.

As a simple example here's what the decision making array looks like for a single ship decision between an island (resource) search and resource collection moves, followed by the final decision:

Resource collection

The final new feature was ensuring that goods are loaded by the computer player on arrival at the resource tile. The amount of goods loaded is sufficient to meet the contract initial delivery requirements.

These screenshots show a ship arriving at a clay resource and the three pottery goods loaded for delivery to the Kingdom island.

That completes the first part of automated decision-making for contracts. The second part will add "resource delivery" moves into the overall strategy, i.e. taking the loaded goods to the Kingdom island.

Pull request / commits

The code changes for the new features can be found here:
https://github.com/miniature-tiger/archipelago/pull/98

---

Proof of Work Carried Out

This is the full url to my github account:
https://github.com/miniature-tiger

---

Contact / Contribute

That is all for this update. If you have any queries please drop them in the comments or contact me on discord.

Have fun!