Farm_RL (WIP)

Reinforcement learning project that creates and solves optimisation problems in the form of farming games.

Usage

To clone the repo and fetch the required dependencies:

git clone https://github.com/FinlaySanders/Farm_RL.git
cd Farm_RL
uv sync

To train a model (will be saved in 'models'):

uv run ppo.py

To train with Weights and Biases logging:

uv run ppo.py --track --wandb-project-name Farm_RL

To visualise a model in 'models':

uv run show.py --model example.pth

Demo V1

Agent has partial observations (within the grey square) - decisions are made using an attention mechanism to weight crops, CNN for pathfinding, and a LSTM for memory.

Screen.Recording.2025-01-18.at.2.09.31.PM.mov

Demo V0

Agent has complete observations - decisions are made with a CNN.

Screen.Recording.2024-11-14.at.15.39.40.mov