Learning from Simulation, Racing in Reality: Sim2Real Methods for Autonomous Racing ...
Reinforcement Learning (RL) methods have been successfully demonstrated in robotic tasks, however, their application to continuous state-action systems with fast dynamics is challenging. In this work, we investigate RL solutions for the autonomous racing problem on the ORCA miniature race car platfo...
Main Author: | |
---|---|
Format: | Text |
Language: | English |
Published: |
ETH Zurich
2020
|
Subjects: | |
Online Access: | https://dx.doi.org/10.3929/ethz-b-000486540 http://hdl.handle.net/20.500.11850/486540 |
id |
ftdatacite:10.3929/ethz-b-000486540 |
---|---|
record_format |
openpolar |
spelling |
ftdatacite:10.3929/ethz-b-000486540 2024-04-28T08:35:14+00:00 Learning from Simulation, Racing in Reality: Sim2Real Methods for Autonomous Racing ... Chisari, Eugenio 2020 application/pdf https://dx.doi.org/10.3929/ethz-b-000486540 http://hdl.handle.net/20.500.11850/486540 en eng ETH Zurich info:eu-repo/semantics/openAccess http://rightsstatements.org/page/InC-NC/1.0/ article-journal Text Master Thesis ScholarlyArticle 2020 ftdatacite https://doi.org/10.3929/ethz-b-000486540 2024-04-02T12:34:54Z Reinforcement Learning (RL) methods have been successfully demonstrated in robotic tasks, however, their application to continuous state-action systems with fast dynamics is challenging. In this work, we investigate RL solutions for the autonomous racing problem on the ORCA miniature race car platform. When training a deep neural network policy using RL methods only using simulations, we observe poor performance, due to model mismatch also known as reality gap. We propose three different methods to reduce this gap, first we propose a policy regularization in the policy optimization step, second, we use model randomization. These two methods allow learning a policy that can race the car without any real environment interactions. Our third method improves this policy, by running the RL algorithm online while driving the car. The achieved performance on the ORCA platform is comparable to that achieved previously by a model-based controller, in terms of lap time, and improved with respect to track constraint ... Text Orca DataCite Metadata Store (German National Library of Science and Technology) |
institution |
Open Polar |
collection |
DataCite Metadata Store (German National Library of Science and Technology) |
op_collection_id |
ftdatacite |
language |
English |
description |
Reinforcement Learning (RL) methods have been successfully demonstrated in robotic tasks, however, their application to continuous state-action systems with fast dynamics is challenging. In this work, we investigate RL solutions for the autonomous racing problem on the ORCA miniature race car platform. When training a deep neural network policy using RL methods only using simulations, we observe poor performance, due to model mismatch also known as reality gap. We propose three different methods to reduce this gap, first we propose a policy regularization in the policy optimization step, second, we use model randomization. These two methods allow learning a policy that can race the car without any real environment interactions. Our third method improves this policy, by running the RL algorithm online while driving the car. The achieved performance on the ORCA platform is comparable to that achieved previously by a model-based controller, in terms of lap time, and improved with respect to track constraint ... |
format |
Text |
author |
Chisari, Eugenio |
spellingShingle |
Chisari, Eugenio Learning from Simulation, Racing in Reality: Sim2Real Methods for Autonomous Racing ... |
author_facet |
Chisari, Eugenio |
author_sort |
Chisari, Eugenio |
title |
Learning from Simulation, Racing in Reality: Sim2Real Methods for Autonomous Racing ... |
title_short |
Learning from Simulation, Racing in Reality: Sim2Real Methods for Autonomous Racing ... |
title_full |
Learning from Simulation, Racing in Reality: Sim2Real Methods for Autonomous Racing ... |
title_fullStr |
Learning from Simulation, Racing in Reality: Sim2Real Methods for Autonomous Racing ... |
title_full_unstemmed |
Learning from Simulation, Racing in Reality: Sim2Real Methods for Autonomous Racing ... |
title_sort |
learning from simulation, racing in reality: sim2real methods for autonomous racing ... |
publisher |
ETH Zurich |
publishDate |
2020 |
url |
https://dx.doi.org/10.3929/ethz-b-000486540 http://hdl.handle.net/20.500.11850/486540 |
genre |
Orca |
genre_facet |
Orca |
op_rights |
info:eu-repo/semantics/openAccess http://rightsstatements.org/page/InC-NC/1.0/ |
op_doi |
https://doi.org/10.3929/ethz-b-000486540 |
_version_ |
1797567383229956096 |