Jose Gibson
← All Projects

CocSim — Clash of Clans RL Simulation

Python simulation of Clash of Clans game mechanics with separate RL agents for building placement and troop deployment, trained to optimise base/attack strategies.

PythonPyTorch / RL training loop (`train.py`)Custom environment (`cocsim_env.py`)Separate building and troop agentsJSON-based game state (`buildings.json`, `troops.json`)
Restricting agents to respect game-rule constraints (troop/bDefining a reward signal rich enough to encourage strategic

Portfolio Highlights

  • Built a multi-agent RL simulation of Clash of Clans game mechanics in Python, with separate policy networks for building placement and troop deployment trained under game-rule constraints.

Snapshot

  • Period: April 2025
  • Source: `github.com/josegibson/cocsim` (private)
  • Domain: Reinforcement learning, game simulation, multi-agent systems
  • Status: Research experiment

Stack

  • Python
  • PyTorch / RL training loop (`train.py`)
  • Custom environment (`cocsim_env.py`)
  • Separate building and troop agents
  • JSON-based game state (`buildings.json`, `troops.json`)
  • Simulator visualizer

What I Built

  • Game environment simulating CoC mechanics: building placement, troop spawning, combat resolution.
  • `BuildingAgent` and `TroopAgent` as separate policy networks with constrained action spaces (allotted counts enforced by design).
  • Training pipeline with environment resets, reward shaping, and logging.
  • Visualizer for stepping through simulation state.
  • Hurdle log documenting design decisions around action-space constraints.