Learning high-speed flight in the wild

Quadrotors are agile. Unlike most other machines, they can traverse extremely complex environments at high speeds. To date, only expert human pilots have been able to fully exploit their capabilities. Autonomous operation with onboard sensing and computation has been limited to low speeds. State-of-...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Science robotics Ročník 6; číslo 59; s. eabg5810
Hlavní autoři: Loquercio, Antonio, Kaufmann, Elia, Ranftl, René, Müller, Matthias, Koltun, Vladlen, Scaramuzza, Davide
Médium: Journal Article
Jazyk:angličtina
Vydáno: United States 06.10.2021
ISSN:2470-9476, 2470-9476
On-line přístup:Zjistit podrobnosti o přístupu
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Abstract Quadrotors are agile. Unlike most other machines, they can traverse extremely complex environments at high speeds. To date, only expert human pilots have been able to fully exploit their capabilities. Autonomous operation with onboard sensing and computation has been limited to low speeds. State-of-the-art methods generally separate the navigation problem into subtasks: sensing, mapping, and planning. Although this approach has proven successful at low speeds, the separation it builds upon can be problematic for high-speed navigation in cluttered environments. The subtasks are executed sequentially, leading to increased processing latency and a compounding of errors through the pipeline. Here, we propose an end-to-end approach that can autonomously fly quadrotors through complex natural and human-made environments at high speeds with purely onboard sensing and computation. The key principle is to directly map noisy sensory observations to collision-free trajectories in a receding-horizon fashion. This direct mapping drastically reduces processing latency and increases robustness to noisy and incomplete perception. The sensorimotor mapping is performed by a convolutional network that is trained exclusively in simulation via privileged learning: imitating an expert with access to privileged information. By simulating realistic sensor noise, our approach achieves zero-shot transfer from simulation to challenging real-world environments that were never experienced during training: dense forests, snow-covered terrain, derailed trains, and collapsed buildings. Our work demonstrates that end-to-end policies trained in simulation enable high-speed autonomous flight through challenging environments, outperforming traditional obstacle avoidance pipelines.
AbstractList Quadrotors are agile. Unlike most other machines, they can traverse extremely complex environments at high speeds. To date, only expert human pilots have been able to fully exploit their capabilities. Autonomous operation with onboard sensing and computation has been limited to low speeds. State-of-the-art methods generally separate the navigation problem into subtasks: sensing, mapping, and planning. Although this approach has proven successful at low speeds, the separation it builds upon can be problematic for high-speed navigation in cluttered environments. The subtasks are executed sequentially, leading to increased processing latency and a compounding of errors through the pipeline. Here, we propose an end-to-end approach that can autonomously fly quadrotors through complex natural and human-made environments at high speeds with purely onboard sensing and computation. The key principle is to directly map noisy sensory observations to collision-free trajectories in a receding-horizon fashion. This direct mapping drastically reduces processing latency and increases robustness to noisy and incomplete perception. The sensorimotor mapping is performed by a convolutional network that is trained exclusively in simulation via privileged learning: imitating an expert with access to privileged information. By simulating realistic sensor noise, our approach achieves zero-shot transfer from simulation to challenging real-world environments that were never experienced during training: dense forests, snow-covered terrain, derailed trains, and collapsed buildings. Our work demonstrates that end-to-end policies trained in simulation enable high-speed autonomous flight through challenging environments, outperforming traditional obstacle avoidance pipelines.
Quadrotors are agile. Unlike most other machines, they can traverse extremely complex environments at high speeds. To date, only expert human pilots have been able to fully exploit their capabilities. Autonomous operation with onboard sensing and computation has been limited to low speeds. State-of-the-art methods generally separate the navigation problem into subtasks: sensing, mapping, and planning. Although this approach has proven successful at low speeds, the separation it builds upon can be problematic for high-speed navigation in cluttered environments. The subtasks are executed sequentially, leading to increased processing latency and a compounding of errors through the pipeline. Here, we propose an end-to-end approach that can autonomously fly quadrotors through complex natural and human-made environments at high speeds with purely onboard sensing and computation. The key principle is to directly map noisy sensory observations to collision-free trajectories in a receding-horizon fashion. This direct mapping drastically reduces processing latency and increases robustness to noisy and incomplete perception. The sensorimotor mapping is performed by a convolutional network that is trained exclusively in simulation via privileged learning: imitating an expert with access to privileged information. By simulating realistic sensor noise, our approach achieves zero-shot transfer from simulation to challenging real-world environments that were never experienced during training: dense forests, snow-covered terrain, derailed trains, and collapsed buildings. Our work demonstrates that end-to-end policies trained in simulation enable high-speed autonomous flight through challenging environments, outperforming traditional obstacle avoidance pipelines.Quadrotors are agile. Unlike most other machines, they can traverse extremely complex environments at high speeds. To date, only expert human pilots have been able to fully exploit their capabilities. Autonomous operation with onboard sensing and computation has been limited to low speeds. State-of-the-art methods generally separate the navigation problem into subtasks: sensing, mapping, and planning. Although this approach has proven successful at low speeds, the separation it builds upon can be problematic for high-speed navigation in cluttered environments. The subtasks are executed sequentially, leading to increased processing latency and a compounding of errors through the pipeline. Here, we propose an end-to-end approach that can autonomously fly quadrotors through complex natural and human-made environments at high speeds with purely onboard sensing and computation. The key principle is to directly map noisy sensory observations to collision-free trajectories in a receding-horizon fashion. This direct mapping drastically reduces processing latency and increases robustness to noisy and incomplete perception. The sensorimotor mapping is performed by a convolutional network that is trained exclusively in simulation via privileged learning: imitating an expert with access to privileged information. By simulating realistic sensor noise, our approach achieves zero-shot transfer from simulation to challenging real-world environments that were never experienced during training: dense forests, snow-covered terrain, derailed trains, and collapsed buildings. Our work demonstrates that end-to-end policies trained in simulation enable high-speed autonomous flight through challenging environments, outperforming traditional obstacle avoidance pipelines.
Author Ranftl, René
Kaufmann, Elia
Müller, Matthias
Koltun, Vladlen
Scaramuzza, Davide
Loquercio, Antonio
Author_xml – sequence: 1
  givenname: Antonio
  orcidid: 0000-0002-8410-3933
  surname: Loquercio
  fullname: Loquercio, Antonio
  organization: University of Zurich, Zürich, Switzerland
– sequence: 2
  givenname: Elia
  orcidid: 0000-0001-6094-5901
  surname: Kaufmann
  fullname: Kaufmann, Elia
  organization: University of Zurich, Zürich, Switzerland
– sequence: 3
  givenname: René
  orcidid: 0000-0003-4158-2759
  surname: Ranftl
  fullname: Ranftl, René
  organization: Intel, Munich, Germany
– sequence: 4
  givenname: Matthias
  orcidid: 0000-0001-5249-8734
  surname: Müller
  fullname: Müller, Matthias
  organization: Intel, Munich, Germany
– sequence: 5
  givenname: Vladlen
  orcidid: 0000-0003-0858-0970
  surname: Koltun
  fullname: Koltun, Vladlen
  organization: Intel, Santa Clara, CA, USA
– sequence: 6
  givenname: Davide
  orcidid: 0000-0002-3831-6778
  surname: Scaramuzza
  fullname: Scaramuzza, Davide
  organization: University of Zurich, Zürich, Switzerland
BackLink https://www.ncbi.nlm.nih.gov/pubmed/34613820$$D View this record in MEDLINE/PubMed
BookMark eNpNj81KxDAYRYOMOOM4TyBIV-KmY74kTdOlDOMPFNzouiRf0jbSP5sW8e0tOIKrexaHC-eSrLq-c4RcA90DMHkf0I-96SePYa9NlSigZ2TDRErjTKRy9Y_XZBfCB6UUUsmlYBdkzYUErhjdkNvc6bHzXRXVvqrjMDhno7JZeIp8F021i758Y6_Ieamb4Han3ZL3x-Pb4TnOX59eDg95jJwmU4wgrXIlYJmWiMZxyIDqREuDNmVUZEKhRSuUTrnSGVLBAUSJFBhfqgzbkrvf32HsP2cXpqL1AV3T6M71cyhYopYOqRK5qDcndTats8Uw-laP38VfG_sBlvRUkw
CitedBy_id crossref_primary_10_1007_s10846_024_02091_6
crossref_primary_10_1016_j_ifacol_2022_07_595
crossref_primary_10_1109_JIOT_2024_3367451
crossref_primary_10_1109_LRA_2023_3281261
crossref_primary_10_1109_TAI_2024_3366871
crossref_primary_10_1016_j_compag_2025_110802
crossref_primary_10_1126_scirobotics_abl8419
crossref_primary_10_1109_LRA_2022_3207545
crossref_primary_10_1016_j_engappai_2025_112090
crossref_primary_10_1038_s42256_025_01050_6
crossref_primary_10_1109_LRA_2022_3176449
crossref_primary_10_1126_scirobotics_ado6187
crossref_primary_10_1109_LRA_2024_3522778
crossref_primary_10_1126_scirobotics_ads5033
crossref_primary_10_1109_TITS_2024_3375893
crossref_primary_10_1109_TIV_2022_3163315
crossref_primary_10_1002_rob_22523
crossref_primary_10_3390_drones7120690
crossref_primary_10_7746_jkros_2024_19_1_117
crossref_primary_10_3390_drones7030192
crossref_primary_10_1109_JSEN_2025_3596433
crossref_primary_10_1109_TNNLS_2024_3525264
crossref_primary_10_1126_scirobotics_adf6991
crossref_primary_10_1109_TRO_2024_3400838
crossref_primary_10_1109_LRA_2022_3181755
crossref_primary_10_1109_LRA_2025_3560842
crossref_primary_10_1371_journal_pone_0264471
crossref_primary_10_1177_02783649231218720
crossref_primary_10_3390_drones8030072
crossref_primary_10_1016_j_ast_2024_109812
crossref_primary_10_3390_s21227436
crossref_primary_10_1109_TIV_2024_3352613
crossref_primary_10_3390_drones7110651
crossref_primary_10_3390_s22249830
crossref_primary_10_1016_j_eswa_2022_116742
crossref_primary_10_1109_TIM_2021_3135544
crossref_primary_10_1007_s10846_023_02042_7
crossref_primary_10_1177_01423312241228886
crossref_primary_10_3390_drones7120704
crossref_primary_10_1109_LRA_2025_3595077
crossref_primary_10_1038_s41467_024_46967_5
crossref_primary_10_1109_TCDS_2024_3519319
crossref_primary_10_1109_LRA_2024_3426374
crossref_primary_10_1126_scirobotics_abm6597
crossref_primary_10_1016_j_ifacol_2023_01_146
crossref_primary_10_1038_s41467_024_54451_3
crossref_primary_10_3389_frobt_2023_1285412
crossref_primary_10_3390_drones7100609
crossref_primary_10_1016_j_eja_2024_127477
crossref_primary_10_1109_TRO_2023_3315710
crossref_primary_10_1109_LRA_2024_3397540
crossref_primary_10_1109_TCST_2023_3297744
crossref_primary_10_1109_LRA_2022_3194682
crossref_primary_10_1007_s41315_025_00457_z
crossref_primary_10_1109_LRA_2024_3357399
crossref_primary_10_1038_s41598_023_36775_0
crossref_primary_10_1002_rob_22460
crossref_primary_10_1109_LRA_2025_3580318
crossref_primary_10_3390_drones8120782
crossref_primary_10_1109_LRA_2024_3371288
crossref_primary_10_3390_aerospace10050418
crossref_primary_10_1002_aisy_202400916
crossref_primary_10_1109_LRA_2025_3562004
crossref_primary_10_1109_TRO_2024_3508140
crossref_primary_10_1109_LRA_2023_3314350
crossref_primary_10_1109_LRA_2025_3606383
crossref_primary_10_1177_02783649231169803
crossref_primary_10_1109_LRA_2022_3150497
crossref_primary_10_3390_s23167206
crossref_primary_10_1109_OJVT_2024_3402129
crossref_primary_10_1109_LRA_2025_3592101
crossref_primary_10_1016_j_knosys_2024_112124
crossref_primary_10_1109_LRA_2025_3548504
crossref_primary_10_1049_ipr2_13079
crossref_primary_10_1109_TVT_2023_3278097
crossref_primary_10_1016_j_trpro_2025_03_100
crossref_primary_10_1109_TRO_2024_3428433
crossref_primary_10_1109_LRA_2024_3399589
crossref_primary_10_1038_s41586_023_06419_4
crossref_primary_10_1007_s11071_024_10029_8
crossref_primary_10_1109_LRA_2025_3553785
crossref_primary_10_1002_aisy_202200113
crossref_primary_10_1109_TRO_2024_3522187
crossref_primary_10_1038_s42256_025_01048_0
crossref_primary_10_1109_TNNLS_2025_3564184
crossref_primary_10_1109_TRO_2023_3334630
crossref_primary_10_1109_ACCESS_2024_3421661
crossref_primary_10_1109_TASE_2024_3432405
crossref_primary_10_1109_LRA_2024_3518078
crossref_primary_10_1109_TIE_2025_3558030
crossref_primary_10_1109_LRA_2025_3547306
crossref_primary_10_1109_LRA_2024_3355632
crossref_primary_10_1016_j_est_2023_107677
crossref_primary_10_1146_annurev_control_030323_022510
crossref_primary_10_1109_LRA_2025_3606806
crossref_primary_10_1109_LRA_2025_3606805
crossref_primary_10_1016_j_birob_2025_100253
crossref_primary_10_1088_2634_4386_ac7ee0
crossref_primary_10_1126_scirobotics_abl6259
crossref_primary_10_1109_TNNLS_2023_3331370
crossref_primary_10_1109_ACCESS_2024_3440183
crossref_primary_10_1109_JIOT_2024_3363036
crossref_primary_10_1109_LRA_2025_3573167
crossref_primary_10_1109_TAES_2025_3535856
crossref_primary_10_1007_s10462_025_11348_x
crossref_primary_10_1177_02783649251353181
crossref_primary_10_3390_robotics11050094
crossref_primary_10_1109_LRA_2022_3146897
crossref_primary_10_1126_scirobotics_adi9641
crossref_primary_10_1109_ACCESS_2023_3330230
crossref_primary_10_1109_TASE_2025_3606896
crossref_primary_10_1109_JIOT_2024_3450967
crossref_primary_10_1109_LRA_2023_3295655
crossref_primary_10_1016_j_robot_2023_104565
crossref_primary_10_1109_TPAMI_2024_3395484
crossref_primary_10_3390_s24196258
crossref_primary_10_1109_TCCN_2024_3358545
crossref_primary_10_1109_TRO_2025_3526102
crossref_primary_10_1109_LRA_2023_3235678
crossref_primary_10_1145_3703453
crossref_primary_10_3390_drones7020144
crossref_primary_10_1109_LRA_2024_3386053
crossref_primary_10_1109_TIE_2023_3323750
crossref_primary_10_1126_scirobotics_add5139
ContentType Journal Article
DBID NPM
7X8
DOI 10.1126/scirobotics.abg5810
DatabaseName PubMed
MEDLINE - Academic
DatabaseTitle PubMed
MEDLINE - Academic
DatabaseTitleList PubMed
MEDLINE - Academic
Database_xml – sequence: 1
  dbid: NPM
  name: PubMed
  url: http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=PubMed
  sourceTypes: Index Database
– sequence: 2
  dbid: 7X8
  name: MEDLINE - Academic
  url: https://search.proquest.com/medline
  sourceTypes: Aggregation Database
DeliveryMethod no_fulltext_linktorsrc
Discipline Engineering
EISSN 2470-9476
ExternalDocumentID 34613820
Genre Research Support, Non-U.S. Gov't
Journal Article
GroupedDBID 0R~
ABJNI
ACGFS
AJGZS
ALMA_UNASSIGNED_HOLDINGS
ARCSS
BKF
EBS
NPM
O9-
SJN
7X8
ID FETCH-LOGICAL-c305t-c16d8ef1cf7fccbe31910a5a6bcd7204948cdcd48a738a9c043114fc0123126b2
IEDL.DBID 7X8
ISICitedReferencesCount 209
ISICitedReferencesURI http://www.webofscience.com/api/gateway?GWVersion=2&SrcApp=Summon&SrcAuth=ProQuest&DestLinkType=CitingArticles&DestApp=WOS_CPL&KeyUT=000705912200001&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D
ISSN 2470-9476
IngestDate Sun Nov 09 11:02:39 EST 2025
Wed Feb 19 02:27:49 EST 2025
IsPeerReviewed true
IsScholarly true
Issue 59
Language English
LinkModel DirectLink
MergedId FETCHMERGED-LOGICAL-c305t-c16d8ef1cf7fccbe31910a5a6bcd7204948cdcd48a738a9c043114fc0123126b2
Notes ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ORCID 0000-0002-8410-3933
0000-0002-3831-6778
0000-0001-5249-8734
0000-0003-0858-0970
0000-0001-6094-5901
0000-0003-4158-2759
PMID 34613820
PQID 2580016856
PQPubID 23479
ParticipantIDs proquest_miscellaneous_2580016856
pubmed_primary_34613820
PublicationCentury 2000
PublicationDate 2021-10-06
PublicationDateYYYYMMDD 2021-10-06
PublicationDate_xml – month: 10
  year: 2021
  text: 2021-10-06
  day: 06
PublicationDecade 2020
PublicationPlace United States
PublicationPlace_xml – name: United States
PublicationTitle Science robotics
PublicationTitleAlternate Sci Robot
PublicationYear 2021
SSID ssj0001763642
Score 2.6176298
Snippet Quadrotors are agile. Unlike most other machines, they can traverse extremely complex environments at high speeds. To date, only expert human pilots have been...
SourceID proquest
pubmed
SourceType Aggregation Database
Index Database
StartPage eabg5810
Title Learning high-speed flight in the wild
URI https://www.ncbi.nlm.nih.gov/pubmed/34613820
https://www.proquest.com/docview/2580016856
Volume 6
WOSCitedRecordID wos000705912200001&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D
hasFullText
inHoldings 1
isFullTextHit
isPrint
link http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwpV1LSwMxEB7UetCD70d9sYJ4i-4jm01OImLxYulBobeSzaMIslu71d_vzDalXgTBS24LYXYm82Vm8n0AV7GILRciZjopFeOYAJj0XjGrlM-z3BlteSs2UfT7cjhUg1Bwa8JY5eJMbA9qWxuqkd-muSR4InNxN_lgpBpF3dUgobEKnQyhDI10FUO5rLFg8IhWPyflRcwUL0QgHgrvZqZ1WRMb8o0ux7lM4t9hZptuetv_3egObAWgGd3PPWMXVly1B5s_6Af34TqQq44jIi1mzQQzWeTf6boevVURQsMIkbQ9gNfe48vDEwu6Ccxg9M6YSYSVzifGF96Y0mGUJbHOtSiNJU0axaWxxnKpi0xqZYhfJ-HeELxCw5TpIaxVdeWOIbKCnprKLBOGGr9OE5-YwkujUKmzievC5cIII_RLajboytWfzWhphi4czS05mswJNEYZ-kSG0OPkD1-fwkZKYyTUsxdn0PEYle4c1s3X7K2ZXrQ_HNf-4Pkbx92z6w
linkProvider ProQuest
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=Learning+high-speed+flight+in+the+wild&rft.jtitle=Science+robotics&rft.au=Loquercio%2C+Antonio&rft.au=Kaufmann%2C+Elia&rft.au=Ranftl%2C+Ren%C3%A9&rft.au=M%C3%BCller%2C+Matthias&rft.date=2021-10-06&rft.eissn=2470-9476&rft.volume=6&rft.issue=59&rft.spage=eabg5810&rft_id=info:doi/10.1126%2Fscirobotics.abg5810&rft_id=info%3Apmid%2F34613820&rft_id=info%3Apmid%2F34613820&rft.externalDocID=34613820
thumbnail_l http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=2470-9476&client=summon
thumbnail_m http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=2470-9476&client=summon
thumbnail_s http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=2470-9476&client=summon