Dual Imitation Learning from Observations (DILO)

Official PyTorch implementation for [A Dual Approach to Imitation Learning from Observations with Offline Datasets) for CoRL 2024. DILO allows for imitating expert observation (action-free) trajectories using suboptimal data.

How to run the code

Install dependencies

Create an empty conda environment and follow the commands below.

conda env create -f environment.yml   
conda activate DILO   

Example training code

Locomotion

python train_dilo.py --env_name=hopper-random-v2 --config=configs/mujoco_config.py --maximizer=smoothed_chi --grad=full  --expert_trajectories=200 --batch_size 1024 --seed=0

Kitchen and Adroit

python train_dilo.py --env_name=hammer-cloned-v0 --config=configs/mujoco_config.py --maximizer=smoothed_chi --grad=full  --expert_trajectories=200 --batch_size 1024 --seed=0

Acknowledgement

This repository builds on the IQL(https://github.com/ikostrikov/implicit_q_learning) codebases. Please make sure to cite them as well when using this code.