This is an example package that provides expert reinforcement learning policies that can be run on the TriFinger robot cluster. You can use it as base for your own package when submitting to the robot cluster. The documentation for the TriFinger RL Datasets provides more information on how to get an account for and submit to the robot cluster.
To install the package run with python 3.8 in the root directory of the repository (we recommend doing this in a virtual environment):
pip install --upgrade pip # make sure the most recent version of pip is installed
pip install .
The package contains two expert policies that were used to record the expert datasets for the Push and Lift tasks in the TriFinger RL datasets.
For the push task:
$ python3 -m trifinger_rl_datasets.evaluate_sim push trifinger_rl_example.example.TorchPushPolicy --n-episodes=3 -v
For the lift task:
$ python3 -m trifinger_rl_datasets.evaluate_sim lift trifinger_rl_example.example.TorchLiftPolicy --n-episodes=3 -v
The policy classes are implemented in trifinger_rl_example/example.py
. The corresponding torch
models are in trifinger_rl_example/policies
and are installed as package_data so they can be loaded
at runtime (see setup.cfg
).
All training checkpoints of the expert policies are available here. They can be
used with this repository by swapping them with the files in the policies
subdirectory.
For more information, please see the software documentation for the TriFinger RL datasets.
The expert policies were introduced in the paper "Benchmarking Offline Reinforcement Learning on Real-Robot Hardware":
@inproceedings{
guertler2023benchmarking,
title={Benchmarking Offline Reinforcement Learning on Real-Robot Hardware},
author={Nico G{\"u}rtler and Sebastian Blaes and Pavel Kolev and Felix Widmaier and Manuel Wuthrich and Stefan Bauer and Bernhard Sch{\"o}lkopf and Georg Martius},
booktitle={The Eleventh International Conference on Learning Representations },
year={2023},
url={https://openreview.net/forum?id=3k5CUGDLNdd}
}
The training pipeline for the expert policies was based on the code of the paper "Transferring dexterous manipulation from GPU simulation to a remote real-world trifinger" by Allshire et al..