A Python implementation of DDPG and PID. Both algorithms are used to solve the control problem of OpenAI Gym's Lunar Lander environment with continuous action space.

The project is a part of the course 02465 Introduction to Reinforcement Learning and Control at DTU.

• DDPG.py contains the classes nessesary for DDPG: OUActionNoise,ReplayBuffer, CriticNetwork, ActorNetwork and Agent
• Train.py contains the train function for DDPG.
• Plots.py contains all nessesary plot functions.
• utils.py contains save methods.
• Grid_Search is the main script for DDPG. It runs a Random Grid Search to tune alpha, beta, tau and gamma for DDPG.
• PID.pyis the main function for PID .
• gym and irlc contains functions and classes nessesary to run PID.

´gym´ and ´irlc´ is a copy of modules from https://gitlab.gbar.dtu.dk/02465material/02465students.git

The PID implementation is inspired by: https://gitlab.gbar.dtu.dk/02465material/02465students.git

The DDPG implementation is inspired by: https://github.com/philtabor/Youtube-Code-Repository/tree/master/ReinforcementLearning/PolicyGradient/DDPG/pytorch/lunar-lander