A model-free method to learn multiple skills in parallel on modular robots
Abstract Legged robots are well-suited for deployment in unstructured environments but require a unique control scheme specific for their design. As controllers optimised in simulation do not transfer well to the real world (the infamous sim-to-real gap), methods enabling quick learning in the real...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2024-07-01
|
| Series: | Nature Communications |
| Online Access: | https://doi.org/10.1038/s41467-024-50131-4 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832594637433339904 |
|---|---|
| author | Fuda van Diggelen Nicolas Cambier Eliseo Ferrante A. E. Eiben |
| author_facet | Fuda van Diggelen Nicolas Cambier Eliseo Ferrante A. E. Eiben |
| author_sort | Fuda van Diggelen |
| collection | DOAJ |
| description | Abstract Legged robots are well-suited for deployment in unstructured environments but require a unique control scheme specific for their design. As controllers optimised in simulation do not transfer well to the real world (the infamous sim-to-real gap), methods enabling quick learning in the real world, without any assumptions on the specific robot model and its dynamics, are necessary. In this paper, we present a generic method based on Central Pattern Generators, that enables the acquisition of basic locomotion skills in parallel, through very few trials. The novelty of our approach, underpinned by a mathematical analysis of the controller model, is to search for good initial states, instead of optimising connection weights. Empirical validation in six different robot morphologies demonstrates that our method enables robots to learn primary locomotion skills in less than 15 minutes in the real world. In the end, we showcase our skills in a targeted locomotion experiment. |
| format | Article |
| id | doaj-art-9a73092b0cea4d7f8bee50036a7259a5 |
| institution | Kabale University |
| issn | 2041-1723 |
| language | English |
| publishDate | 2024-07-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | Nature Communications |
| spelling | doaj-art-9a73092b0cea4d7f8bee50036a7259a52025-01-19T12:29:31ZengNature PortfolioNature Communications2041-17232024-07-0115111310.1038/s41467-024-50131-4A model-free method to learn multiple skills in parallel on modular robotsFuda van Diggelen0Nicolas Cambier1Eliseo Ferrante2A. E. Eiben3Department of Computer Science, Vrije Universiteit AmsterdamDepartment of Computer Science, Vrije Universiteit AmsterdamDepartment of Computer Science, Vrije Universiteit AmsterdamDepartment of Computer Science, Vrije Universiteit AmsterdamAbstract Legged robots are well-suited for deployment in unstructured environments but require a unique control scheme specific for their design. As controllers optimised in simulation do not transfer well to the real world (the infamous sim-to-real gap), methods enabling quick learning in the real world, without any assumptions on the specific robot model and its dynamics, are necessary. In this paper, we present a generic method based on Central Pattern Generators, that enables the acquisition of basic locomotion skills in parallel, through very few trials. The novelty of our approach, underpinned by a mathematical analysis of the controller model, is to search for good initial states, instead of optimising connection weights. Empirical validation in six different robot morphologies demonstrates that our method enables robots to learn primary locomotion skills in less than 15 minutes in the real world. In the end, we showcase our skills in a targeted locomotion experiment.https://doi.org/10.1038/s41467-024-50131-4 |
| spellingShingle | Fuda van Diggelen Nicolas Cambier Eliseo Ferrante A. E. Eiben A model-free method to learn multiple skills in parallel on modular robots Nature Communications |
| title | A model-free method to learn multiple skills in parallel on modular robots |
| title_full | A model-free method to learn multiple skills in parallel on modular robots |
| title_fullStr | A model-free method to learn multiple skills in parallel on modular robots |
| title_full_unstemmed | A model-free method to learn multiple skills in parallel on modular robots |
| title_short | A model-free method to learn multiple skills in parallel on modular robots |
| title_sort | model free method to learn multiple skills in parallel on modular robots |
| url | https://doi.org/10.1038/s41467-024-50131-4 |
| work_keys_str_mv | AT fudavandiggelen amodelfreemethodtolearnmultipleskillsinparallelonmodularrobots AT nicolascambier amodelfreemethodtolearnmultipleskillsinparallelonmodularrobots AT eliseoferrante amodelfreemethodtolearnmultipleskillsinparallelonmodularrobots AT aeeiben amodelfreemethodtolearnmultipleskillsinparallelonmodularrobots AT fudavandiggelen modelfreemethodtolearnmultipleskillsinparallelonmodularrobots AT nicolascambier modelfreemethodtolearnmultipleskillsinparallelonmodularrobots AT eliseoferrante modelfreemethodtolearnmultipleskillsinparallelonmodularrobots AT aeeiben modelfreemethodtolearnmultipleskillsinparallelonmodularrobots |