Keywords
Machine Learning, Disturbances, Trial and Error

Technologies
Python, Pytorch, Gym

This project combines reinforcement learning with classical control to design and stabilise a simulated quadcopter system. The hybrid approach ensures both adaptability and reliability under varying conditions. Simulation environments such as Gym and PyBullet were used to train and test the system, with a focus on learning control policies for stable flight. Extensive testing was carried out under disturbances including wind, reduced engine power, and increased mass to evaluate robustness. A key finding was the identification of a critical point where performance stops improving and may degrade. This highlights important limitations and challenges when applying these methods to real-world, unpredictable environments.