Imitation-Learning-PyTorch

Basic Behavioural Cloning and DAgger Implementation in PyTorch

Behavioural Cloning:

• Define your policy network model in model.py.
• Get appropriate states from environment. Here I am creating random episodes during training.
• Extract the expert action here from a .txt file or a pickle file or some function of states.
• Run python imitation_algos.py.

Requirements:

• numpy
• pytorch
• tensorboardX
• tqdm

To Do

• Implement DAgger
• Make it compatible with OpenAI gym environments
• Add supporting algorithms to train further with Reinforcement Learning
• Comments in the code :P

For tensorboard:

tensorboard --logdir=runs --host localhost --port 8088