Reinforcement Learning for Quadrotor trajectory control

In this project, multiple reinforcement learning algorithms were tested for learning quadrotor take off.

The environment used is a custom created environment that uses pybullet as a simulator.

The tested algorithms are DDPG, SAC, PPO, TRPO.

Running

For best algorithm results SAC, run the following in the project root folder

python3 SAC.py

To run other algorithms, inside the root folder run the following in the project root folder

python3 <algo>.py

Where <algo> is one of the following: pid,DDPG, SAC, PPO, TRPO.

Dependencies

• numpy
• scipy
• matplotlib
• pybullet==3.2.5
• gym==0.21.0
• torch==1.13.0
• stable_baselines3==1.8.0
• sb3_contrib==1.8.0 for TRPO implementation

This project is implemented as a course project for the 'Machine Learning In Robotics' course.