This is codebase for paper Policy Contrastive Imitation Learning. This codebase is build upon ROT .

In this repo, we provide intro to

• get start
• download expert demo
• train agents

• Install Mujoco based on the instructions given here.

• Installing the following libraries

  apt install libosmesa6-dev libgl1-mesa-glx libglfw3
  

• installing conda environments

   conda env create -f conda_env.yml
   cd <root_to_pcil_codebase>
   conda activate pcil 
  

Download expert demonstrations, weights and environment libraries [link] The link contains the following:

• The expert demonstrations for all tasks in the paper.
• The weight files for the expert (DrQ-v2) and behavior cloning (BC).
• The supporting libraries for environments (Gym-Robotics, metaworld) in the paper.
• Extract the files provided in the link
  • set the path/to/dir portion of the root_dir path variable in cfgs/config.yaml to the path of the PCIL repository.
  • place the expert_demos and weights folders in ${root_dir}/PCIL.

We use online logger [WANDB]:(https://wandb.ai), you can customize your own logger in logger.py.

We provide an example script:

python train.py agent=pc suite=dmc  device_id=0 obs_type=features  suite/dmc_task=hopper_stand  num_demos=10  

You can customize more parameters in config files in ./cfgs