20230504_安信证券_汽车零部件行业证券研究报告：AI大模型在自动驾驶中的应用_32页.pdf

1 2023 05 04 AI-A 688326-W 160-A Wind%1M 3M 12M-8.0-8.9 13.5-7.3-12.5 16.2 SAC S1450520040002 SAC S1450523040001 2023-04-21 2023-03-15 2023 2.0 2022-12-12 2022-10-29 2022-05-17 Transformer Transformer Attention 2021 Transformer CNN CNN Transformer RNN Transformer Transformer 2D BEV AI 2023 Corner case NGP BEV Transformer 1 raw-data 2018 2D 4D AI Transformer AI Dojo 1.4 GPU AI 30%AI 170,-13%-3%7%17%27%37%47%2022-05 2022-09 2022-12 2023-04 300/2 1.AI.4 1.1.4 1.1.1.1.DNN.5 1.1.1.2.CNN.5 1.1.1.3.RNN.7 1.1.2.Transformer.8 1.2.BEV+Transformer.10 1.2.1.10 1.2.2.Transformer BEV.12 2.AI.15 2.1.NGP AI.15 2.2.AI NGP.17 2.2.1.BEV+transformer“”.18 2.2.2.“”.21 2.2.3.22 3.AI.23 3.1.+.23 3.2.26 4.28 4.1.28 4.2.28 4.3.29 5.29 1.4 2.5 3.DNN.5 4.6 5.CNN DNN/.6 6.CNN.7 7.CNN.7 8.RNN.8 9.Transformer.8 10.Vision Transformer.8 11.9 12.-.9 13.transformer.10 14.-.11 15.11 16.12 0YAZxOmRnOnMpRnOmRqPnMaQdNaQnPrRoMpMlOoOmRfQrRtO8OnMtQvPoNpMwMqRoM/3 17.12 18.BEV.13 19.BEV.13 20.LSS.14 21.Transformer BEV.14 22.BEV.15 23.BEV BEV.15 24.16 25.Corner case.17 26.Corner case.17 27.18 28.BEV.19 29.BEV.20 30.BEV.20 31.XNet.20 32.BEV.21 33.BEV.22 34.23 35.23 36.24 37.24 38.25 39.26 40.26 41.27 42.28 43.28 44.29 1 2023.16 2.18 3.19 4.21 5.27 /4 1.AI 1.1.1.70 2010 CNN RNN Transformer 1 x x=a b c a b c 2 w=w1 w2 w3 3 z xwT+z0,4 xwT+b /5 2.1.1.1.1.DNN DNN 3.DNN 1.1.1.2.CNN DNN CNN CNN /6 4.CNN DNN CNN DNN 8 8 64 CNN 3 3 Pooling 6 6 2 2 Maxpooling 1/2 5.CNN DNN/7 6.CNN CNN CNN CNN CNN 7.CNN 1.1.1.3.RNN RNN CNN CNN RNN RNN RNN DNN RNN RNN RNN 1 RNN 2 RNN RNN /8 8.RNN 1.1.2.Transformer Transformer NLP 2020 CV 2017 NIPS Attention is all you need Transformer a=(a1,a2,a3,.,ai,.)Transformer token token Transformer NLP NLP 2020 VIT Vision Transformer Patch Transformer Transformer CV 9.Transformer 10.Vision Transformer An Image is Worth 16x16 Words:Transformers for Image Recognition at Scale Transformer attention attention X /9 attention 11.Transformer Encoder Decoder Multi-Head Self-Attention Multi-Head Cross-Attention Transformer Q Query K Key V Value Q K V(,)=()1 2 3 Q K V Q K/V 12.-/10 13.transformer Attention Is All You Need Transformer 2021 Transformer CNN CNN transformer RNN Transformer 1.2.BEV+Transformer 1.2.1./11 14.-1 Mobileye 2 100Tops 3 2020 15.SoC 1 /12 2 500-1000Tops 100Tops 16.3/4D 3/4D 500-1000Tops 300-400Tops transformer BEV 17.1.2.2.Transformer BEV BEV Birds Eye View BEV BEV BEV BEV IPM 2D 2D IPM IPM BEV 2D BEV /13 18.BEV AI Day BEV BEV 1 2D Backbone feature 2+BEV 3 4D 4 19.BEV BEV 2020 LSS BEV 2D LSS CNN 3D BEV /14 20.LSS Lift,splat,shoot:Encoding images from arbitrary camera rigs byimplicitly unprojecting to 3d Transformer BEV 2D BEV Query Key/Value 2D BEV BEV 3D Query 2D 2021 AI Day Transformer BEV BEV 21.Transformer BEV AI Day /15 22.BEV AI Day 23.BEV BEV AI Day 2.AI 2.1.NGP AI L2 TJA HWA APA 2022 L2 L2+585.99 61.66%29.40%2021-2022 L2+2021 7 FSD BetaV9 2022-2023 /16 1 2023 2021 1 26 NGP 2022 9 27 NGP 2023 3 31 G9 P7i Max NGP 2023H2 NGP 2024+2020 10 NOP 2023 1 NOP+Beta 2021 NOA 2022 AD Max 2.0 Lidar AD Pro 2.0 NOA 2023Q2 NOA 2023 NOA 100 M5 2023Q2 NCA 5 2023Q3 NCA 15 2023Q4 NCA 45 S 2022 9 NCA 2023 3 21 NCA 2023 4 L7 5 NOA 2023 NOA NGP NOA NCA NOP 2023 2023+100%90%71%24.Corner case NOA Corner case AI+/17 25.Corner case Vehicle 26.Corner case Vehicle 2.2.AI NGP 1 2018 task 3 2020-2021 image image BEVtransformer BEV task 4 2022 BEV 2020 8 2021 7 FSD Beta V9.0 2022 11 FSD Beta V11 Transformer 4 /18 27.CSDN 2.2.1.BEV+transformer+SLAM 1 2 3 14 Waymo 1TB 8h 1 3 2()()()()2022 9 2021 31 2022 19 /19 license+license 1000 100-500 3 license+license+BEV+transformer 3D 2D 3D 3D 2021 AI Day BEV BEV 10-20 BEV 28.BEV Vehicle BEV/BEV BEV 10-20 BEV /20 29.BEV 30.BEV AI Day BEV+Transformer 1024 G9 XNGP XNet BEV 31.XNet /21 4 2022/10/24 AI Day XNGP 2023/01/05 HAOMO AI DAY NOH 2023/01/28 CEO AD Max 3.0 2023/03/23 2023 ADS 2.0 2023/3/22 DeepRoute-Driver 3.0 2023/1/11 PonyClassic PonyPro PonyUltra Momenta 2023/01/12 IM TECH DAY IM AD NOA 2022/2/28 TA PILOT 4.0 AI Day 2.2.2.BEV 2022 AI Day Occupancy Network voxel voxel 0/1 voxel voxel 1 voxel voxel 0 voxel 32.BEV AI Day AI Day /22 33.BEV AI Day 4D HW4.0 4D 3D 4D 4D MEMS 3000-9000 4D 1000-2000 2.2.3.BEV Occupancy BEV end-to-end raw-data/AI-DRL DRL /23 34.CSDN 2016 End to End Learning for Self-Driving Cars CNN Steering command CNN/CNN 35.End to End Learning for Self-Driving Cars Occupancy Occupancy Occupancy 3.AI 3.1.+Transformer 1 CNN Transformer CEO AI DAY 1 Transformer CNN /24 36.AI Day raw-data 2018 4 1 1(2018)2 2 2(2019)3D 3 3(2020)BEV 4 4(2021)4D 500 37.AI Day /25 45,000 2,000 16.7 AI Day 4D Clip 100%98%38.AI Day 1 1%1 1 2 3 Simulation World Creator 1 2 3 4 /26 39.40.AI Day AI Day DRIVE Sim Replicator DRIVE Sim DRIVE Replicator 1 DRIVE Sim RTX 2 DRIVE Sim 3D AI corner case AI corner case 3.2.Transformer CEO AI Day Attention Transformer CNN 100 AI Dojo 1.4 GPU AI AI Day 14 10 GPU Dojo AI 30%1 2 CPU GPU/AI AI /27 AI 3 41.AI Day IDC 1 2 20%3 2022 8 AI CEO 60 60000TFLOPs 170 10100 5 Dojo 2021.08 180/2022.08 60/170 419-2022.11-2023.01 81/20%2025 120/MANA OASIS 2023.01 67/62%110%8 GPU 2000 GPU 2022.06-2022.09 20/2022.12 400/AIDC 2022.01 491/28 AI Day 42.43.IDC IDC 4.4.1.Tier1 2016 2022 25.7 83.07%17.22%IPU01 IPU04 IPU01 IPU02 L1 L2 Xavier/Orin IPU03 IPU04 L2+L2+4.2.L4 L4 ADAS L2 ADCU L2+/L3 HPC L4 3 0%5%10%15%20%25%30%35%40%车企/主机厂和车企传统一级供应商 互联网大厂和芯片厂商AI 科技公司自动驾驶独角兽0%10%20%30%40%50%60%3 周以内 4-6 周 7-12 周 13-18 周 19-24 周 超过14 周车企/主机厂和车企传统一级供应商 互联网大厂和芯片厂商AI 科技公司自动驾驶独角兽/29 44.4.3.AI 2021 2022 L2+L4 SD5223 L2+16 TOPS SOC 8M IFC/5V5R/10V5R SD5226 L4 7nm AI 400 TOPS CPU 300K+DMIPs 2022 290 370 590 290 AI OTA AI 5.2022-2023 Transformer 2020 8 2021 7 NOA L2+/30 6 300 10%6 300-10%10%6 300 10%A 6 300 B 6 300 /31 19 33 518026 638 3 200080 2 15 100034 前沿报告库是中国新经济产业咨询报告共享平台。行业范围涵盖新一代信息技术、5G、物联网、新能源、新材料、新消费、大健康、大数据、智能制造等新兴领域。为企事业单位、科研院所、投融资机构等提供研究和决策参考。扫一扫免费获取海量报告