Experiments

Benchmark Study of Meta-Policy Search Methods

To reproduce the experimental results, run the following scripts in our code repository:

python experiments/all_envs_eval/ppo_run_all.py
python experiments/all_envs_eval/ppo_run_sawyer.py
python experiments/all_envs_eval/trpo_run_all.py
python experiments/all_envs_eval/trpo_run_sawyer.py
python experiments/all_envs_eval/vpg_run_all.py
python experiments/all_envs_eval/vpg_run_sawyer.py

The experimental data is available under the following link:

https://www.dropbox.com/sh/lh6cbhu9yhiex08/AAB1hJ6jpjVYD6pJma4uTEgDa?dl=0

Gradient Variance of Curvature Estimators (DICE vs. LVC)

To reproduce the experimental results, run the following scripts in our code repository:

python experiments/gradient_variance/run_sweep.py

The experimental data is available under the following link:

https://www.dropbox.com/sh/5dlnd7smeulif4e/AABbo_hpmI7lpur62d0HlF9-a?dl=0

Comparison of Initial Sampling Distributions


To reproduce the experimental results, run the following scripts in our code repository:

python experiments/exploration_comparison/trpo_run_sweep.py

The experimental data is available under the following link:

https://www.dropbox.com/sh/qs9iorsxx3qejrk/AABy_BKbBZc26KQUepifvJ_ma?dl=0