Benchmarking Structured Policies and Policy Optimization for Real-World Dexterous Object Manipulation
Dexterous manipulation is a challenging and important problem in robotics. While data-driven methods are a promising approach, current benchmarks require simulation or extensive engineering support due to the sample inefficiency of popular methods. We present benchmarks for the TriFinger system, an open-source robotic platform for dexterous manipulation and the focus of the 2020 Real Robot Challenge. The benchmarked methods, which were successful in the challenge, can be generally described as structured policies, as they combine elements of classical robotics and modern policy optimization. This inclusion of inductive biases facilitates sample efficiency, interpretability, reliability and high performance. The key aspects of this benchmarking is validation of the baselines across both simulation and the real system, thorough ablation study over the core features of each solution, and a retrospective analysis of the challenge as a manipulation benchmark. The code and demo videos for this work can be found on our website (https://sites.google.com/view/benchmark-rrc).
| Author(s): | Niklas Funk and Charles Schaff and Rishabh Madan and Takuma Yoneda and Julen Urain and Joe Watson and Ethan K. Gordon and Felix Widmaier and Stefan Bauer and Siddhartha S. Srinivasa and Tapomayukh Bhattacharjee and Matthew R. Walter and Jan Peters |
| Journal: | IEEE Robotics and Automation Letters (RA-L) |
| Volume: | 7 |
| Number (issue): | 1 |
| Pages: | 478--485 |
| Year: | 2022 |
| Month: | January |
| Bibtex Type: | Article (article) |
| DOI: | 10.1109/LRA.2021.3129139 |
| State: | Published |
| URL: | https://arxiv.org/abs/2105.02087 |
| Electronic Archiving: | grant_archive |
| Links: | |
BibTex
@article{rrc2020_benchmark,
title = {Benchmarking Structured Policies and Policy Optimization for Real-World Dexterous Object Manipulation},
journal = {IEEE Robotics and Automation Letters (RA-L)},
abstract = {Dexterous manipulation is a challenging and important problem in robotics. While data-driven methods are a promising approach, current benchmarks require simulation or extensive engineering support due to the sample inefficiency of popular methods. We present benchmarks for the TriFinger system, an open-source robotic platform for dexterous manipulation and the focus of the 2020 Real Robot Challenge. The benchmarked methods, which were successful in the challenge, can be generally described as structured policies, as they combine elements of classical robotics and modern policy optimization. This inclusion of inductive biases facilitates sample efficiency, interpretability, reliability and high performance. The key aspects of this benchmarking is validation of the baselines across both simulation and the real system, thorough ablation study over the core features of each solution, and a retrospective analysis of the challenge as a manipulation benchmark. The code and demo videos for this work can be found on our website (https://sites.google.com/view/benchmark-rrc).},
volume = {7},
number = {1},
pages = {478--485},
month = jan,
year = {2022},
slug = {rrc2020_benchmark},
author = {Funk, Niklas and Schaff, Charles and Madan, Rishabh and Yoneda, Takuma and Urain, Julen and Watson, Joe and Gordon, Ethan K. and Widmaier, Felix and Bauer, Stefan and Srinivasa, Siddhartha S. and Bhattacharjee, Tapomayukh and Walter, Matthew R. and Peters, Jan},
url = {https://arxiv.org/abs/2105.02087},
month_numeric = {1}
}