🚁 Drone RL + PID Controller + Hybrid RL

Reinforcement Learning for Precise Trajectory Tracking Using PyBullet + PPO

This repository contains a full **Reinforcement Learning ** control system for trajectory-tracking quadcopters.
The environment is built on gym-pybullet-drones, with a custom modified RL environment that teaches a Crazyflie-style drone to fly:

• ✔ Circles
• ✔ Figure-8 paths
• ✔ Four-point patterns
• ✔ “Go-To” single-point moves

---

🎥 Drone Flight Demo (Click to Watch)

train  python .\scripts\trainer.py # make sure you set a task to train [circle, four_points, figure8,goto]
test src/testing/tester.py 

Single-Task & Multi-Task Training

The environment supports:

• Single-task RL (train only circle, only four_points, etc.)
• Sequential training (train circle → load model → train figure8 → …)

Each task has its own set of waypoints and reward shaping.

---

Waypoint-Based Motion

Each trajectory is expressed as a list of waypoints:

The agent receives:

• Current position
• Velocity
• Orientation
• Angular velocity
• Relative vector to current waypoint

---

4. Live Training Plot

A live matplotlib window shows:

• Rolling mean episode reward
• Training stability
• Improvements over time

---