JetsonNanoで動く深層強化学習を使ったラジコン向け自動運転ソフトウェアの紹介

Transcript

1. +FUTPO
   /BOPͰಈ͘ਂ૚ڧԽֶशΛ࢖ͬͨ
   ϥδίϯ޲͚ࣗಈӡసιϑτ΢ΣΞͷ঺հ *P5
   "-(:"/
   प೥*P5ࡇ
   
   
   !NBTBUP@LB
   
2. ͋͞ʂ"*
   3$
   $BSΛ࢝ΊΑ͏ ࢢൢϥδίϯ 4#$ "*ͱΧϝϥͷΈ -&7&-૬౰ͷ૸ߦ ੈքதͰϨʔε͕։࠵͞Ε͍ͯΔ ৄ͘͠͸IUUQNBTBUPLBIBUFOBCMPHDPN
3. "*
   3$
   $BS૸ߦͷ࢓૊Έ 4FMG%SJWJOH
   $JSDVJU ……….. 
   ڭࣔσʔλͷऩू 
   "*ͷֶश 
   ਪ࿦૸ߦ "*Λ࢖͍ೝ஌͔Β൑அ·Ͱ&OE
   UP
   &OEͰ૸ߦ͢Δ ৄ͘͠͸IUUQNBTBUPLBIBUFOBCMPHDPN
4. "*
   3$
   $BSϨʔεͰউͭͨΊʹʂ ϋʔυ΢ΣΞ "*$7੍ޚ ύΠϩοτదੑ ڭࣔσʔλ͕উഊΛେ͖͘࡞༻͢Δ ৄ͘͠͸IUUQNBTBUPLBIBUFOBCMPHDPN
5. ͓खຊͳ͠ʹ૸ΓํΛֶͿਂ૚ڧԽֶश ࢼߦͱใु͔Βྑ͍૸Γͷಛ௃ΛֶͿ 4FMG%SJWJOH
   $JSDVJU ΞΫγϣϯ Environment AI͕σʔλΛ ूΊΔɻ ※ใुͷ૯࿨͕࠷େͱͳΔΑ͏ͳಈ͖ํΛ୳ࡧ͍ͯ͠Δɻ ใु,ঢ়ଶ ৄ͘͠͸IUUQNBTBUPLBIBUFOBCMPHDPN Jetson NanoͰ10෼ʙ20෼Ͱ૸ߦํ๏Λ֫ಘ
6. ࣮૷ͨ͠ιϑτ΢ΣΞͷಛ௃ JETSON NANO SAC VAE Haarnoja, T., Zhou, A., Abbeel, P., and Levine, S. Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor. arXiv preprint arXiv:1801.01290, 2018. ΤοδͰਂ૚ڧԽֶशΛ͢ΔͨΊͷ޻෉ Antonin Raffin, Learning to Drive Smoothly in Minutes Reinforcement Learning on a Small Racing Car , 2019. ৄ͘͠͸IUUQNBTBUPLBIBUFOBCMPHDPN
7. ιϑτͷެ։ͱ൓ڹ (JUIVCͷ3&"%.&͸ӳޠ (PPHMF຋༁ Ͱެ։ ৄ͘͠͸IUUQNBTBUPLBIBUFOBCMPHDPN /7*%*"͞Μͷ
   +FUTPO
   $PNNVOJUZ
   1SPKFDUT IUUQTEFWFMPQFSOWJEJBDPNFNCFEEFEDPNNVOJUZKFUTPOQSPKFDUTSFJOGPSDFNFOU@KFUCPU
8. ໨ࢦͤόʔνϟϧϨʔε ੈքதͷϢʔβ͕ࢀՃͯ͠ΦϯϥΠϯͰରઓ γϛϡϨʔλΛ࢖ͬͨϨʔε ΦϯϥΠϯͰϦΞϧλΠϜରઓ ੈքνϟϯϐΦϯ͸೔ຊਓ
   ৄ͘͠͸IUUQNBTBUPLBIBUFOBCMPHDPN
9. ·ͱΊ "*
   3$
   $BSΛ࢝ΊͯΈ·ͤΜ͔ʁ "*Ͱܭଌ੍ޚΛֶͿ"*
   3$
   $BS ΤοδσόΠεͰਂ૚ڧԽֶश ੈքͷίϛϡχςΟͱܨ͕Ζ͏ ৄ͘͠͸IUUQNBTBUPLBIBUFOBCMPHDPN