Documentation for the TriFinger RL Datasets
This documentation describes the TriFinger RL datasets, how to use them and how to submit reinforcement learning (RL) jobs to the TriFinger cluster. The TriFinger platform was designed for dexterous manipulation and can operate without human supervision due to a bowl-shaped barrier and regular self checks. The TriFinger RL datasets consist of more than 100 hours of trajectories recorded with policies of varying proficiency. Possible use cases for these datasets include offline RL and imitation learning but also learning of dynamics models and representation learning.
For more information on the TriFinger robot and on how to request remote access to a cluster of such robots, visit the TriFinger website.
If you use the TriFinger RL Datasets in your research, please cite the following paper:
@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}
}