ImitationLearning

Assignment - Imitation Learning

The goal of this assignment is to experiment with imitation learning - direct behavior cloning and the DAgger algorithm, on OpenAI Gym's CarRacing-v0 environment.

Instructions

Follow the general instructions to clone the repository.
Specific instructions may be found in the comments in each script.

Installing Dependencies

1. Required Libraries

   pip3 install -r requirements.txt

2. Installing resources for the environment.

   git clone https://github.com/openai/gym.git
   cd gym
   pip install -e .
   pip3 install -e '.[box2d]'

Creating the Expert Policy

Note : This will work only on a local machine

A small dataset of expert behaviour is already present in data/data.pkl.gzip. You are, however encouraged to create your own dataset by running manual.py.

Defining the model

Edit model.py to add data processing, if required and define your agent model.

Training

Train your agent by defining the training pipelines in train.py for -

• Naive Behavioral Cloning
• DAgger

Run the training script :

usage: train.py [-h] [--mode {naive,dagger}]

optional arguments:
  -h, --help            show this help message and exit
  --mode {naive,dagger}, -m {naive,dagger}
                        Sets the training mode. Default : naive

Evaluation

Add the commands for loading and getting predictions from your trained model in eval.py and run the script.
Note : rendering in colab can only be achieved by plotting the state returned with env.render(mode='rgb_array').

usage: eval.py [-h] [--render] [--num_episodes NUM_EPISODES]

    optional arguments:
    -h, --help            show this help message and exit
    --render, -r          To visualise the agent's performance
    --num_episodes NUM_EPISODES, -n NUM_EPISODES
                            Number of episodes to run.