| | --- |
| | license: apache-2.0 |
| | tags: |
| | - reinforcement-learning |
| | - stable-baselines3 |
| | - ppo |
| | - game |
| | - gymnasium |
| | - pygame |
| | library_name: stable-baselines3 |
| | task: reinforcement-learning |
| | --- |
| | |
| | # PixelCopter RL Agent |
| |
|
| |  |
| |
|
| | ## Overview |
| |
|
| | PixelCopter is a side-scrolling reinforcement learning environment inspired by the classic helicopter game. The agent: |
| |
|
| | - Controls a **pixel copter** navigating through a scrolling cavern. |
| | - Avoids colliding with walls that move from right to left. |
| | - Gains positive rewards for surviving timesteps; receives negative rewards on collision. |
| |
|
| | This repository contains: |
| |
|
| | - `train_pixelcopter.py` – Training script |
| | - `test_pixelcopter.py` – Test script with Pygame visualization |
| | - `ppo_pixelcopter_cert.zip` – Pre-trained PPO agent |
| |
|
| | --- |
| |
|
| | ## Environment Details |
| |
|
| | - **Observation Space:** `[copter_y, wall_gap_positions...]` |
| | - **Action Space:** |
| | - `0`: Do nothing (gravity pulls the copter down) |
| | - `1`: Apply lift (move upwards) |
| | - **Rewards:** |
| | - +1 for surviving a timestep |
| | - -5 for collision |
| | - **Medium Difficulty Settings:** |
| | - Gap size: 6 |
| | - Gravity: 0.25 |
| | - Max timesteps per episode: 500 |
| |
|
| | --- |
| |
|
| | ## Installation |
| |
|
| | ```bash |
| | pip install stable-baselines3 gymnasium pygame numpy |
| | |
