Repository files navigation

This repository is a collection of the following reinforcement learning algorithms:

• Policy-Gradient
• Actor-Critic
• Trust Region Policy Optimization
• Generalized Advantage Estimation
• Proximal Policy Optimization

More algorithms will be added on this repository.

In this repository,OpenAI Gym environments such asCartPole-v0,Pendulum-v0, andBipedalWalker-v3 are used. You need to install them before running this repository.

Note: The environment's names could be different depending on the version of OpenAI Gym.

1. Install Python 3.

2. Install the Python packages inrequirements.txt. If you are using a virtual environment for Python package management, you can install all python packages needed by using the following bash command:

   $ pip install -r requirements.txt

3. Install other packages to run OpenAI Gym environments. These are dependent on the development setting of your machine.

4. Install PyTorch. The version of PyTorch should be greater or equal than 1.7.0.

1. Modifyconfig.json as your machine setting.

2. Execute training process bytrain.py. An example of usage fortrain.py are following:

   $ python train.py --model_name=trpo --env_name=BipedalWalker-v3

   The following bash command will help you:

   $ python train.py -h

3. You can run your pre-trained agents by executingrun.py. The usage for runningrun.py is similar to that oftrain.py. You can also check the help message by the following bash bash command:

   $ python run.py -h

• The CUDA usage is provided now.
• Modified some errors in GAE and PPO.
• Modified some errors about horizon was corrected.

• Find the errors of the Actor-Critic
• Implement ACER
• Search other environments to running the algorithms

• An explaination of TRPO line search:link
• Additional stability method for PPO value function:link