MBIRL - Model Based Inverse Reinforcement Learning

Simulation with ground truth keypoint predictions

Generate expert demonstrations

1. python mbirl/generate_expert_demo.py
2. Check the data and visualizations of the demonstration in 'mbirl/traj_data'

Run Our Method

1. python mbirl/experiments/run_model_based_irl.py
2. Check the trajectories predicted during training in model_data/placing/<cost_type>

Plot the losses, evaluate our method

1. jupyter notebook
2. Access the notebook in the browser in 'mbirl/experiments/plot_mbirl_training_and_eval.ipynb'

Simulation with learned keypoint representation and dynamics

COMING SOON