Autotune

1. Installation

Install https://github.com/vitchyr/multiworld and pip install -e . (Similar with https://github.com/vitchyr/rlkit). Pytorch version: 1.1.0 with or without gpu is ok.

Create your personal rlkit/launchers/config.py like: https://drive.google.com/file/d/1Nl-olFC-h5MRG91QoDirEoZb0iF5AmYj/view?usp=sharing.

2. The Simulated Experiment

python examples/autotune_all/rig_sac_push_autotune.py
python examples/autotune_all/debug_autotune_r_size.py

3. The Hardware Experiment

git fetch original ros:ros
git checkout ros
python examples/rig_sac_automatic/elfin_reach_auto.py

References

The algorithm of Auto-tuning is based on the following paper:

Hyperparameter Auto-tuning in Self-Supervised Robotic Learning. Jiancong Huang, Juan Rojas, Matthieu Zimmer, Hongmin Wu, Yisheng Guan, and Paul Weng. arXiv preprint, 2021.