Quadcopter RL Environment

This repository provides a custom OpenAI Gymnasium environment for simulating and training reinforcement learning (RL) agents to control a quadcopter. It includes a reward function, environment implementation, and a sample notebook for training with the TD3 algorithm using Stable Baselines3.

Features

• Custom Environment: QuadcopterEnv simulates a 12-dimensional quadcopter state and 4-dimensional action space.
• Reward Function: Flexible quadratic reward function for state and action penalties.
• TD3 Training Example: Jupyter notebook (td3.ipynb) demonstrates training and evaluation of a TD3 agent.
• Logging: Training logs are saved in the logs/ directory for analysis.

File Overview

• quad_copter.py: Defines the QuadcopterEnv class, a Gymnasium-compatible environment for quadcopter control.
• reward_func.py: Contains the quadcopter_reward function, a quadratic cost-based reward for RL.
• td3.ipynb: Jupyter notebook for training and evaluating a TD3 agent on the custom environment.
• logs/: Directory for training logs and monitor files.

Getting Started

Prerequisites

• Python 3.8+
• gymnasium
• stable-baselines3
• numpy
• pandas (for log analysis)

Install dependencies:

pip install gymnasium stable-baselines3 numpy pandas

Usage

1. Custom Environment: Use QuadcopterEnv from quad_copter.py in your RL experiments.
2. Reward Function: Import and use quadcopter_reward for custom reward shaping.
3. Training: Run the td3.ipynb notebook to train and evaluate a TD3 agent.

Example

from quad_copter import QuadcopterEnv
import numpy as np

env = QuadcopterEnv()
obs, _ = env.reset()
for _ in range(100):
    action = env.action_space.sample()
    obs, reward, done, info = env.step(action)
    if done:
        break

Notes

• The environment state is a 12D vector: position, angles, velocities, and angular velocities.
• The action is a 4D vector, typically representing motor commands.
• The reward penalizes deviation from the goal state and large actions.