Rapidly exploring learning trees
Inverse Reinforcement Learning (IRL) for path planning enables robots to learn cost functions for difficult tasks from demonstration, instead of hard-coding them. However, IRL methods face practical limitations that stem from the need to repeat costly planning procedures. In this paper, we propose R...
| Main Authors: | , , |
|---|---|
| Format: | Conference item |
| Published: |
IEEE
2017
|
| _version_ | 1797087516814213120 |
|---|---|
| author | Shiarlis, K Messias, J Whiteson, S |
| author_facet | Shiarlis, K Messias, J Whiteson, S |
| author_sort | Shiarlis, K |
| collection | OXFORD |
| description | Inverse Reinforcement Learning (IRL) for path planning enables robots to learn cost functions for difficult tasks from demonstration, instead of hard-coding them. However, IRL methods face practical limitations that stem from the need to repeat costly planning procedures. In this paper, we propose Rapidly Exploring Learning Trees (RLT∗ ), which learns the cost functions of Optimal Rapidly Exploring Random Trees (RRT∗ ) from demonstration, thereby making inverse learning methods applicable to more complex tasks. Our approach extends Maximum Margin Planning to work with RRT∗ cost functions. Furthermore, we propose a caching scheme that greatly reduces the computational cost of this approach. Experimental results on simulated and real-robot data from a social navigation scenario show that RLT∗ achieves better performance at lower computational cost than existing methods. We also successfully deploy control policies learned with RLT∗ on a real telepresence robot. |
| first_indexed | 2024-03-07T02:36:47Z |
| format | Conference item |
| id | oxford-uuid:a90c2aa7-1821-40f2-b1fb-5b1cc15493c3 |
| institution | University of Oxford |
| last_indexed | 2024-03-07T02:36:47Z |
| publishDate | 2017 |
| publisher | IEEE |
| record_format | dspace |
| spelling | oxford-uuid:a90c2aa7-1821-40f2-b1fb-5b1cc15493c32022-03-27T03:05:49ZRapidly exploring learning treesConference itemhttp://purl.org/coar/resource_type/c_5794uuid:a90c2aa7-1821-40f2-b1fb-5b1cc15493c3Symplectic Elements at OxfordIEEE2017Shiarlis, KMessias, JWhiteson, SInverse Reinforcement Learning (IRL) for path planning enables robots to learn cost functions for difficult tasks from demonstration, instead of hard-coding them. However, IRL methods face practical limitations that stem from the need to repeat costly planning procedures. In this paper, we propose Rapidly Exploring Learning Trees (RLT∗ ), which learns the cost functions of Optimal Rapidly Exploring Random Trees (RRT∗ ) from demonstration, thereby making inverse learning methods applicable to more complex tasks. Our approach extends Maximum Margin Planning to work with RRT∗ cost functions. Furthermore, we propose a caching scheme that greatly reduces the computational cost of this approach. Experimental results on simulated and real-robot data from a social navigation scenario show that RLT∗ achieves better performance at lower computational cost than existing methods. We also successfully deploy control policies learned with RLT∗ on a real telepresence robot. |
| spellingShingle | Shiarlis, K Messias, J Whiteson, S Rapidly exploring learning trees |
| title | Rapidly exploring learning trees |
| title_full | Rapidly exploring learning trees |
| title_fullStr | Rapidly exploring learning trees |
| title_full_unstemmed | Rapidly exploring learning trees |
| title_short | Rapidly exploring learning trees |
| title_sort | rapidly exploring learning trees |
| work_keys_str_mv | AT shiarlisk rapidlyexploringlearningtrees AT messiasj rapidlyexploringlearningtrees AT whitesons rapidlyexploringlearningtrees |