Online vs. Offline Adaptive Domain Randomization Benchmark

Gabriele Tiboni*, Karol Arndt, Giuseppe Averta, Ville Kyrki, Tatiana Tommasi

*Tämän työn vastaava kirjoittaja

Tutkimustuotos: Artikkeli kirjassa/konferenssijulkaisussaConference article in proceedingsScientificvertaisarvioitu

1 Sitaatiot (Scopus)

Abstrakti

Physics simulators have shown great promise for conveniently learning reinforcement learning policies in safe, unconstrained environments. However, transferring the acquired knowledge to the real world can be challenging due to the reality gap. To this end, several methods have been recently proposed to automatically tune simulator parameters with posterior distributions given real data, for use with domain randomization at training time. These approaches have been shown to work for various robotic tasks under different settings and assumptions. Nevertheless, existing literature lacks a thorough comparison of existing adaptive domain randomization methods with respect to transfer performance and real-data efficiency. This work presents an open benchmark for both offline and online methods (SimOpt, BayRn, DROID, DROPO), to investigate current limitations on multiple settings and tasks. We found that online methods are limited by the quality of the currently learned policy for the next iteration, while offline methods may sometimes fail when replaying trajectories in simulation with open-loop commands. The code used is publicly available at https://github.com/gabrieletiboni/adr-benchmark.

AlkuperäiskieliEnglanti
OtsikkoHuman-Friendly Robotics 2022 - HFR
Alaotsikko15th International Workshop on Human-Friendly Robotics
ToimittajatPablo Borja, Cosimo Della Santina, Luka Peternel, Elena Torta
KustantajaSpringer
Sivut158-173
Sivumäärä16
ISBN (elektroninen)978-3-031-22731-8
ISBN (painettu)978-3-031-22730-1
DOI - pysyväislinkit
TilaJulkaistu - 2023
OKM-julkaisutyyppiA4 Artikkeli konferenssijulkaisussa
TapahtumaInternational Workshop on Human-Friendly Robotics - Delft, Alankomaat
Kesto: 22 syysk. 202223 syysk. 2022

Julkaisusarja

NimiSpringer Proceedings in Advanced Robotics
Vuosikerta26
ISSN (painettu)2511-1256
ISSN (elektroninen)2511-1264

Workshop

WorkshopInternational Workshop on Human-Friendly Robotics
LyhennettäHFR
Maa/AlueAlankomaat
KaupunkiDelft
Ajanjakso22/09/202223/09/2022

Sormenjälki

Sukella tutkimusaiheisiin 'Online vs. Offline Adaptive Domain Randomization Benchmark'. Ne muodostavat yhdessä ainutlaatuisen sormenjäljen.

Siteeraa tätä