A competitive hex based restoration strategy game designed for reinforcement learning ✨
Bloomward is a custom strategy game project built around flowers, spirits, seasons, and spreading corruption.
It is being designed as both:
🌱 a fun and strategic game for human play
🤖 a structured environment for reinforcement learning experiments
Players compete on a shared hex board, placing flowers, forming combos, activating spirits, and trying to survive the pressure of corruption as it spreads inward toward the Sacred Tree.
Bloomward explores how a game can be:
- 🌸 strategically interesting for players
- 🧠 suitable for AI and reinforcement learning
- 🍃 themed around restoration, balance, and environmental pressure
Instead of using a standard benchmark game, this project creates a custom game environment from scratch.
The board has three main layers:
-
Sacred Core 🌳
The center of the board, containing the Sacred Tree -
Fertile Soil 🌱
The main playable area where flowers can be placed -
Corrupt Soil 🖤
The outer edge, where corruption begins and spreads inward
Current flower types include:
- 🌻 Sunflower
- 🌷 Tulip
- 🌼 Blossom
Players draw and place flowers on valid fertile tiles.
Flower placements can form combos that awaken spirits.
- 🔺 3 matching flowers in a triangle activates a spirit
- 🌸 9 flower cluster activates a stronger restoration effect
Spirits help defend, heal, or stabilize the board.
Corruption is one of the main pressure mechanics in the game.
- starts at the outer edge
- spreads inward over time
- only spreads onto empty fertile soil
- reduces space and threatens the Sacred Tree
This creates urgency and forces smarter decisions.
Seasonal changes keep the game dynamic and can affect how pressure builds across the board.
Bloomward is also being developed as a reinforcement learning environment.
The long term goal is to model the game so that an agent can learn through interaction with the board by:
- observing game states
- choosing valid actions
- receiving rewards
- adapting to corruption, seasons, and an opponent
This makes Bloomward a useful testbed for game AI and sequential decision making.
This project currently focuses on:
- 📘 formalising the game rules
- 🎮 building a digital prototype
- 🧩 modelling the environment for reinforcement learning
- 🤖 preparing for future agent training
The first prototype includes:
- hexagonal board representation
- two player turn system
- flower placement
- combo detection
- spirit activation
- corruption spread
- season cycling
- win and loss checking
For now, the priority is functionality over polish.
Planned next steps include:
- improving the prototype
- refining rule consistency
- defining state and action spaces
- designing reward functions
- testing baseline agents
- training reinforcement learning agents
R. Solomons
🎓 Honours Project 2026
🏫 University of the Western Cape
Bloomward is currently being developed for academic and research purposes.
A formal license and fuller documentation can be added later.
🌸 Bloomward • Strategy • Restoration • Reinforcement Learning 🤖