Variable Impedance Skill Learning for Contact-Rich Manipulation
(2022) In IEEE Robotics and Automation Letters 7(3). p.8391-8398- Abstract
- Contact-rich manipulation tasks remain a hard problem in robotics that requires interaction with unstructured environments. Reinforcement Learning (RL) is one potential solution to such problems, as it has been successfully demonstrated on complex continuous control tasks. Nevertheless, current state-of-the-art methods require policy training in simulation to prevent undesired behavior and later domain transfer even for simple skills involving contact. In this paper, we address the problem of learning contact-rich manipulation policies by extending an existing skill-based RL framework with a variable impedance action space. Our method leverages a small set of suboptimal demonstration trajectories and learns from both position, but also... (More)
- Contact-rich manipulation tasks remain a hard problem in robotics that requires interaction with unstructured environments. Reinforcement Learning (RL) is one potential solution to such problems, as it has been successfully demonstrated on complex continuous control tasks. Nevertheless, current state-of-the-art methods require policy training in simulation to prevent undesired behavior and later domain transfer even for simple skills involving contact. In this paper, we address the problem of learning contact-rich manipulation policies by extending an existing skill-based RL framework with a variable impedance action space. Our method leverages a small set of suboptimal demonstration trajectories and learns from both position, but also crucially impedance-space information. We evaluate our method on a number of peg-in-hole task variants with a Franka Panda arm and demonstrate that learning variable impedance actions for RL in Cartesian space can be deployed directly on the real robot, without resorting to learning in simulation. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/0d10c436-e705-4ed3-8748-87f92ff3588e
- author
- Yang, Quantao ; Dürr, Alexander LU ; Topp, Elin Anna LU ; Stork, Johannes and Stoyanov, Todor
- organization
- publishing date
- 2022-06-30
- type
- Contribution to journal
- publication status
- published
- subject
- keywords
- Machine learning for Robot Control, Reinforcement Learning, Variable Impedance Control
- in
- IEEE Robotics and Automation Letters
- volume
- 7
- issue
- 3
- pages
- 8 pages
- publisher
- IEEE - Institute of Electrical and Electronics Engineers Inc.
- external identifiers
-
- scopus:85133737407
- ISSN
- 2377-3766
- DOI
- 10.1109/LRA.2022.3187276
- project
- Reinforcement Learning in Continuous Spaces with Interactively Acquired Knowledge-based Models
- Robot Skill Learning based on Interactively Acquired Knowledge-based Models
- language
- English
- LU publication?
- yes
- id
- 0d10c436-e705-4ed3-8748-87f92ff3588e
- date added to LUP
- 2022-09-06 13:39:50
- date last changed
- 2023-11-21 11:03:00
@article{0d10c436-e705-4ed3-8748-87f92ff3588e, abstract = {{Contact-rich manipulation tasks remain a hard problem in robotics that requires interaction with unstructured environments. Reinforcement Learning (RL) is one potential solution to such problems, as it has been successfully demonstrated on complex continuous control tasks. Nevertheless, current state-of-the-art methods require policy training in simulation to prevent undesired behavior and later domain transfer even for simple skills involving contact. In this paper, we address the problem of learning contact-rich manipulation policies by extending an existing skill-based RL framework with a variable impedance action space. Our method leverages a small set of suboptimal demonstration trajectories and learns from both position, but also crucially impedance-space information. We evaluate our method on a number of peg-in-hole task variants with a Franka Panda arm and demonstrate that learning variable impedance actions for RL in Cartesian space can be deployed directly on the real robot, without resorting to learning in simulation.}}, author = {{Yang, Quantao and Dürr, Alexander and Topp, Elin Anna and Stork, Johannes and Stoyanov, Todor}}, issn = {{2377-3766}}, keywords = {{Machine learning for Robot Control; Reinforcement Learning; Variable Impedance Control}}, language = {{eng}}, month = {{06}}, number = {{3}}, pages = {{8391--8398}}, publisher = {{IEEE - Institute of Electrical and Electronics Engineers Inc.}}, series = {{IEEE Robotics and Automation Letters}}, title = {{Variable Impedance Skill Learning for Contact-Rich Manipulation}}, url = {{http://dx.doi.org/10.1109/LRA.2022.3187276}}, doi = {{10.1109/LRA.2022.3187276}}, volume = {{7}}, year = {{2022}}, }