Bee Foraging RL Simulation

This repository contains a Reinforcement Learning (RL) project in which a bee collects pollen from flowers and returns it to a hive. The agent uses Q-learning to discover efficient foraging strategies under various vision constraints.

Features

• Grid Environment: A 6×6 board with a hive and multiple flowers.
• Health Mechanic: Bee loses health with each move; dying ends the episode.
• Pollen Delivery: Collect pollen from flowers and earn a large reward upon hive return.
• Vision Modes:
  1. Full Vision: Bee knows its exact (x,y) location and carrying status.
  2. 3×3 Partial Vision: Bee only sees a local 3×3 window around its position.
  3. Single-Tile Vision: Bee sees only the tile it stands on.

Key Files

• bee_foraging_simulation.py: Main Python script running Q-learning with adjustable vision modes.
• bee.png, flower1.png, flower2.png, flower3.png: Images for rendering the bee and flowers.
• Dependencies:
  • pygame for rendering
  • matplotlib for real-time plotting
  • python 3.7+

How to Run

1. Install dependencies: pip install pygame matplotlib
2. Launch the simulation: bash python bee_foraging_simulation.py
3. Adjust parameters (episodes, learning rate, vision mode) in the script to experiment with different settings.

Learning Logic

Q-Learning

State can be:

• (x, y, carrying_flag) for full vision, or
• (local_3x3_view, carrying_flag) for partial vision, or
• (single_tile_view, carrying_flag) for single-tile vision.

Actions: Move up/down/left/right (or turn/move if orientation-based).

Rewards:

• Small penalty for each step.
• Positive for picking up pollen.
• Large bonus for delivering pollen to the hive.
• Penalty for running out of health.

---

Observations

Even with partial or single-tile vision, the bee can eventually learn routes back to the hive in a small environment, because Q-learning’s table effectively encodes the path from each state to the hive.

Feel free to fork or modify the project for your own experiments. Happy foraging!