Implementation of various policy based reinforcement learning algorithms, detailed in Open AI's Spinning Up resource. In order of complexity, these include:
- pg0.py: Use full return to estimate policy grad.
- pg1.py: Use reward-to-go to estimate policy grad.
- pg2.py: Use reward-to-go and value baseline.
- vpg.py: Vanilla Policy Gradient.
- ppo.py: Proximal Policy Optimization.
The implementations are compatible with RL environments provided by the
gymnasium library.
We test the algorithms on the CartPole-v1 environment.