pendulum-v1

Implementing Deep Reinforcement Learning Algorithms in Python for use in the MuJoCo Physics Simulator

Continuous Environment

Pendulum-v1 comparison of 4 DQN variants with explicit torque discretisation, 25 tests across Python 3.10–3.12, and 84.91% coverage.

Training an agent in the gym Pendulum-v1 environment using Actor-Critic algorithm.

Contains Expert Trajectories for various Gym Environments used for State Only Imitation Learning

This project investigates to what extend evolutionary methods such as the Cross Entropy Method and Evolution Strategies can be used to optimize a neural policy compared to the baseline REINFORCE.

Empirical study of PPO hyperparameter robustness under unseen dynamics shifts in Pendulum-v1, using random search, multi-seed evaluation, and zero-shot evaluation under gravity shifts.