Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Hello DeepRacer World
Search
cmtt
October 19, 2019
Technology
0
1k
Hello DeepRacer World
re:Invent 2018で発表されたAWS DeepRacerが、今年大いに盛り上がりました。
DeepRacerの概要と始め方、各レースの様子を紹介します。
cmtt
October 19, 2019
Tweet
Share
More Decks by cmtt
See All by cmtt
re:Invent2019注目トピックとDeepRacerアップデート
cmtt
0
660
Other Decks in Technology
See All in Technology
今年一年で頑張ること / What I will do my best this year
pauli
1
190
When Windows Meets Kubernetes…
pichuang
0
270
知っててうれしい HTTP Cookie を使ったセッション管理について
greendrop
1
120
Zero Data Loss Autonomous Recovery Service サービス概要
oracle4engineer
PRO
1
5k
AIエージェントに脈アリかどうかを分析させてみた
sonoda_mj
2
140
Fearsome File Formats
ange
0
580
機械学習を「社会実装」するということ 2025年版 / Social Implementation of Machine Learning 2025 Version
moepy_stats
3
410
.NET 最新アップデート ~ AI とクラウド時代のアプリモダナイゼーション
chack411
0
160
I could be Wrong!! - Learning from Agile Experts
kawaguti
PRO
8
3k
The future we create with our own MVV
matsukurou
0
1.8k
AWSの生成AIサービス Amazon Bedrock入門!(2025年1月版)
minorun365
PRO
7
410
ヤプリQA課題の見える化
gu3
0
160
Featured
See All Featured
Optimising Largest Contentful Paint
csswizardry
33
3k
How to Ace a Technical Interview
jacobian
276
23k
A better future with KSS
kneath
238
17k
Dealing with People You Can't Stand - Big Design 2015
cassininazir
365
25k
Making Projects Easy
brettharned
116
6k
Facilitating Awesome Meetings
lara
50
6.2k
Unsuck your backbone
ammeep
669
57k
個人開発の失敗を避けるイケてる考え方 / tips for indie hackers
panda_program
98
18k
GitHub's CSS Performance
jonrohan
1030
460k
It's Worth the Effort
3n
183
28k
Building a Scalable Design System with Sketch
lauravandoore
460
33k
CSS Pre-Processors: Stylus, Less & Sass
bermonpainter
356
29k
Transcript
)FMMP%FFQ3BDFS8PSME "84ࣄۀຊ෦ɹαʔϏεάϧʔϓ ߴڮ
"CPVUNF ߴڮ "84ࣄۀຊ෦αʔϏεάϧʔϓ ʮΫϥεϝιουɾϝϯόʔζʯ ϋʔυΣΞʹڵຯ͕͋Δ ιϑτΣΞ͞Μ UXJUUFS!FEZ HJUIVCFEZ
ຊͷ͓
AWS DeepRacer
AWS DeepRacer
͓͠ͳ͕͖ wͳͥ"84͕ϥδίϯΛʁ wػցֶशͱڧԽֶश w%FFQ3BDFSͷ͡Ίํ w4VNNJU$JSDVJUͱ7JSUVBM$JSDVJU w%FFQ3BDFS$IBNQJPOTIJQ$VQJO-BT7FHBT
Φεεϝͳਓ wϨʔε͖ͳਓ wܕϥδίϯ͖ͳਓ wϩϘοτ͖ͳਓ wػցֶशΓ͍͚ͨͲ·ͩͬͯͳ͍ਓ wͪΖΜػցֶशΨν
ͳͥʹAWS͕ϥδίϯΛʁ
ͳͥʹ"84͕ϥδίϯΛʁ w։ൃऀ͚ڧԽֶशϋϯζΦϯ wػցֶशڧԽֶशΛͬͱ։ൃऀʹΊ͍ͨ w͞ΘͬͯɺࢼߦࡨޡͰ͖ΔϓϥοτϑΥʔϜ wΘ͔Γ͍͢ඪɿϨʔεʹࢀՃͯ͠উͭ
"84%FFQ3BDFS
"84%FFQ3BDFS wεέʔϧͷϥδίϯ wڧԽֶशʹΑΔࣗߦ wϚχϡΞϧߦ 8J'J
εϖοΫ
6OEFSUIFIPPE
6OEFSUIFIPPE
6OEFSUIFIPPE
6OEFSUIFIPPE
ϥδίϯ࣮ػͷΞʔΩςΫνϟ wΧϝϥ w0QFO7*/0 w304
ػցֶशͱڧԽֶश
ػցֶशͷछྨ wڭࢣֶ͖श wڭࢣແֶ͠श wڧԽֶश ɹɹˢ%FFQ3BDFS͜Ε http://www0.cs.ucl.ac.uk/staff/d.silver/web/Teaching_files/intro_RL.pdf
ڧԽֶश͕ΘΕ͍ͯΔͱ͜Ζ
ڧԽֶशͷΈ
ใुؔʹྑ͞Λࣔؔ͢ wݴޠQZUIPO marker_1 = 0.1 * track_width marker_2 =
0.25 * track_width marker_3 = 0.5 * track_width reward = 1e-3 if distance_from_center >= 0.0 and distance_from_center <= marker_1: reward = 1 elif distance_from_center <= marker_2: reward = 0.5 elif distance_from_center <= marker_3: reward = 0.1 else: reward = 1e-3
γϛϡϨʔγϣϯͷΞʔΩςΫνϟ
%FFQ3BDFSͷ%-ωοτϫʔΫ
DeepRacerͷίϯιʔϧ
ϚωʔδϝϯτίϯιʔϧͰ%FFQ3BDFS
࠷ॳͷτϨʔχϯά
Ϧιʔεͷ࡞ ඞਢϦιʔεͷ࡞͕ඞཁ ճҎޙෆཁ
Ϟσϧ໊ΛܾΊΔ ໊લΛ͚Δ ͨ͘͞Μ࡞ΔͷͰࣈͰ۠ผ ΦϓγϣϯͰϝϞ
ΞΫγϣϯεϖʔεͷઃఆ
ใुؔ γϯλοΫενΣοΫ ฤूՄೳ αϯϓϧ͔ΒબΜͰ0,
ϋΠύʔύϥϝʔλ w࠷ॳσϑΥϧτͰ0, w܁Γฦ࣌͠ʹΛߋ৽ w-FBSOJOH3BUF w&OUSPQZ
γϛϡϨʔγϣϯͰτϨʔχϯά
γϛϡϨʔγϣϯͰτϨʔχϯά
τϨʔχϯάΛ܁Γฦ͢
τϨʔχϯά ϞσϧΛ࡞ ΞΫγϣϯεϖʔεΛઃఆ τϨʔχϯάίʔεΛબ ใुؔͱϋΠύʔύϥϝʔλΛઃఆ τϨʔχϯά )d) ධՁ
ϞσϧΛίϐʔ 3͔Β܁Γฦ͢
ֶश࣌ؒ w؆୯ͳίʔεͳΒ࣌ؒͰͰ͖Δ͜ͱ͋Δ w݅Λม͑Δͱͳ͔ͳֶ͔श͕ਐߦ͠ͳ͍ wίʔεͷΧʔϒ؇ٸ wߴઃఆͷΞΫγϣϯεϖʔε wਐߦঢ়گΛݟͳ͕Βd)ఔֶशͤ͞Δ͜ͱ
%FFQ3BDFSʹ͔͔Δඅ༻ wτϨʔχϯάͱධՁͷ࣮ߦ࣌ؒʹର͠)͘Β͍ w7JSUVBM$JSDVJUͰͷܭଌʹඅ༻͔͔Βͳ͍ w܁Γฦֶ͠शΛଓ͚ΔͷͰ݁ߏඅ༻͕͔͔Γ·͢ wਖ਼ݴͬͯେਓͷ༡ͼͰ͢ wΫϨδοτ͕ͨΔ8FFLMZ$IBMMFOHFاըΞϦ
wϩάͷࢹ֮Խͷྫ wπʔϧ͕"84༗ࢤ͔Β ڞ༗͞Ε͍ͯΔ ੳ
Virtual Circuit
7JSUVBM$JSDVJU wΦϯϥΠϯͰৗ࣌։࠵ wίʔε݄ସ w֤݄ͷॱҐΛڝ͏ w༏উऀܾউτʔφϝϯτਐग़ݖΛ֫ಘ ɹˠ-BT7FHBTͱSF*OWFOUট
݄ସΘΓͷίʔε Pre Season London Loop Kumo Torakku Empire City
Shanghai Sudu Cumulo Carrera Toronto Turnpike 4݄ 5݄ 6݄ 7݄ 8݄ 9݄ 10݄
τϨʔχϯάίʔεͱαʔΩοτίʔε αʔΩοτίʔε τϨʔχϯάίʔε Toronto Turnpike Toronto Turnpike Training
ࠓ݄'JOBM0OMJOF3BDF
8FFLMZ$IBMMFOHF wຖि͓͕ઃఆ͞ΕͯҐ·Ͱදজ wҬผɺϧʔΩʔɺվળτοϓɺಛఆඵʹ͍ۙ wॱҐͰΫϨδοτఏڙ d
Summit Circuit
4VNNJU$JSDVJU wੈք֤Ͱ։࠵͞ΕΔ"844VNNJU wίʔε͕ઃஔ͞Εɺ࣮ࡍʹΒͤΔ͜ͱ͕Ͱ͖Δ wେਓؾͰ࣌ؒͪβϥ
4VNNJU$JSDVJU wϞσϧɺ࣋ͪࠐΈग़དྷ߹͍ͷͲͪΒͰྑ͍ w࣋ͪ࣌ؒɺ४උɺίʔε w࠷ϥοϓ͕ग़ΔΑ͏͕࣌ؒڐ͢ݶΓτϥΠ wτϥοΫϘε͕ੈΛ͢Δ ίʔεΞτͨ͠Β͢ ɹˢେมͳॏ࿑ಇͰ͢ ײँ
4VNNJUͷίʔε wίʔεछྨɿSF*OWFOU
࣮ࡍʹΒͤΔ
࣮ػϔͷσϓϩΠ w࡞ֶͨ͠शϞσϧΛμϯϩʔυ w64#ϝϞϦʹอଘ wϥδίϯʹଓ ɹɹˠϞσϧΛ࣮ػʹϩʔυɺ࠷దԽ wྃ͢Δͱ8FCը໘ͰબΔΑ͏ʹͳΔ
"844VNNJU5PLZP
"844VNNJU5PLZP
"844VNNJU5PLZP
"844VNNJU5PLZP
"844VNNJU5PLZP
"844VNNJU5PLZP https://www.youtube.com/watch?v=yDGm7u8wrNM
None
7JSUVBMͱͷҧ͍ wJ1BEͰεϩοτϧͷૢ࡞͕Մೳ wػମిͳͲͷݸମࠩΛௐ͢Δ wͳ͔ͳ͔ࢥ͏௨Γʹ Ԡ͠ͳͯ͘େม
࣮ࡍʹΒͤͯΈΔͱ w݈ؾʹΔ࢟ͪΐͬͱײಈ͢Δ wͨͱ͑ྑ͍λΠϜ͕ग़ͳͯ͘
.(.4QFFEXBZ!SF*OWFOU
.(.)BDLFS4QBDF
World Championship for Developpers
$IBNQJPOTIJQ$VQ wՕॴͷ4VNNJU$JSDVJUͷ্Ґ w֤݄ͷ7JSUVBM$JSDVJU༏উऀ w૯߹ϙΠϯτ্Ґ w߹Θ໊͕ͤͯϥεϕΨεͷେձࢀՃ wνϟϯϐϣϯΧοϓΛڝ͍·͢
ܾউϦʔάਐग़ऀ https://aws.amazon.com/jp/blogs/machine-learning/the-aws-deepracer-league-and-countdown-to-the-reinvent-championship-cup-2019/ https://aws.amazon.com/jp/blogs/machine-learning/the-aws-deepracer-league-and-countdown-to-the-reinvent-championship-cup-2019/
"84%FFQ3BDFS57 w%FFQ3BDFSઐ༻ϝσΟΞ w֤ͷ4VNNJU$JSDVJUதܧΠϯλϏϡʔ https://www.youtube.com/watch?v=yDGm7u8wrNM
"84%FFQ3BDFS57 https://aws.amazon.com/deepracer/league
4MBDL͋Γ·͢ wEFFQSBDFSDPNNVOJUZTMBDLDPN ɹɹ໊ ݱࡏ ɻݱࡏট੍ɻӳޠ wEFFQSBDFSKQTMBDLDPN ɹɹ໊ɻຊޠɻࣗ༝ࢀՃ w͝ڵຯ͕͋Εট͠·͢ͷͰ͓͕͚͍ͩ͘͞
+"84'&45"4"11310։࠵ w%FFQ3BDFSίʔεઃஔʂ wϫʔΫγϣοϓ ߦձΛ༧ఆʂ wKGU
·ͱΊ
·ͱΊ w%FFQ3BDFS։ൃऀָ͕͘͠ػցֶशڧԽֶशΛ ֶΔϓϥοτϑΥʔϜ wڧԽֶशͷݪཧͱτϨʔχϯάํ๏ֶ͕Δ wόʔνϟϧϨʔεʹࢀՃͯ͠᛭ୖຏ w࣮ػͷϨʔε·ͨҧͬͨ໘ന͞ wੈքେձࢀՃͷӫ༪Λࢦͤʂ
None